Paper deep dive
An Evaluation Framework for National AI Regulation
Kaushik Sanjay Prabhakar, Tarun Adarsh R S, Amal Dhivyan Gregory, Sreeparvathy Sajeev, Utkarsh Tomar, Avyay M Casheekar
Intelligence
Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 93%
Last extracted: 8/22/2026, 2:13:49 AM
Summary
This paper introduces an evaluation framework for assessing the design and implementation readiness of national AI policy portfolios. It compares the regulatory approaches of China, India, Japan, Singapore, South Korea, the UK, the US, and the EU, focusing on legal instruments, institutional capacity, and lifecycle coverage rather than just prominent laws. The framework scores policies based on risk governance, institutional ability, and protections, providing a traceable comparison of policy content on paper.
Entities (13)
Relation Signals (8)
Evaluation Framework for National AI Regulation → compares → China
confidence 95% · The comparison covers China...
Evaluation Framework for National AI Regulation → compares → European Union
confidence 95% · The European Union is included as a supranational comparator.
Evaluation Framework for National AI Regulation → evaluates → National AI Policy Portfolios
confidence 95% · The framework evaluates a versioned portfolio of official instruments rather than one prominent law or strategy.
AI Act → governs → European Union
confidence 90% · The AI Act entered into force on August 1, 2024... It prohibits specified practices and places obligations on high-risk systems.
Executive Order 14110 → revokedby → Executive Order 14179
confidence 90% · Executive Order 14110 was revoked. Executive Order 14179 then directed a new plan...
Executive Order 14110 → directed → United States
confidence 85% · Executive Order 14110 directed federal action on AI safety and security.
NIST AI Risk Management Framework → maintainedby → United States
confidence 85% · NIST continues to maintain the voluntary AI Risk Management Framework...
DeepSeek → released → V3
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Governments use laws, institutions, funding programs and nonbinding guidance to shape how AI is developed and used. Comparing these national approaches is difficult. A binding rule and a detailed voluntary framework can address the same problem but create different duties. The resources needed to carry them out also differ by jurisdiction. This paper develops an evaluation framework for the documented design and implementation readiness of national AI policy. The comparison covers China, India, Japan, Singapore, South Korea, the United Kingdom and the United States. The European Union is included as a supranational comparator. The framework evaluates a versioned portfolio of official instruments rather than one prominent law or strategy. Its criteria ask whether the portfolio governs serious AI risks and whether responsible institutions can implement its commitments. They examine coverage across the AI lifecycle and the protections available to people affected by AI systems. Public benefit and responsible innovation remain a separate part of the assessment. Each sub-criterion is scored through ordered anchors and tied to the provision that supports the judgment. The protocol also records the source search, missing evidence, included instruments and cutoff date. The result is a traceable comparison of policy content that keeps category differences visible. It evaluates what a portfolio provides on paper. It does not estimate enforcement success or policy outcomes.
Tags
Links
- Source: https://arxiv.org/abs/2608.15417v1
- Canonical: https://arxiv.org/abs/2608.15417v1
Trouble viewing inline? Open PDF directly →
Full Text
242,164 characters extracted from source content.
Expand or collapse full text
An Evaluation Framework for National AI Regulation Kaushik Sanjay Prabhakar, 1 Tarun Adarsh R S, 2 Amal Dhivyan Gregory, 2 Sreeparvathy Sajeev, 2 Utkarsh Tomar, 2 Avyay M Casheekar 3 1 University of Illinois Chicago 2 Vellore Institute of Technology 3 University of Michigan Law School Abstract Governments use laws, institutions, funding programs and nonbinding guidance to shape how AI is developed and used. Comparing these national approaches is dif- ficult. A binding rule and a detailed voluntary frame- work can address the same problem but create differ- ent duties. The resources needed to carry them out also differ by jurisdiction. This paper develops an evalua- tion framework for the documented design and imple- mentation readiness of national AI policy. The compari- son covers China, India, Japan, Singapore, South Korea, the United Kingdom and the United States. The Euro- pean Union is included as a supranational comparator. The framework evaluates a versioned portfolio of offi- cial instruments rather than one prominent law or strat- egy. Its criteria ask whether the portfolio governs serious AI risks and whether responsible institutions can imple- ment its commitments. They examine coverage across the AI lifecycle and the protections available to peo- ple affected by AI systems. Public benefit and respon- sible innovation remain a separate part of the assess- ment. Each sub-criterion is scored through ordered an- chors and tied to the provision that supports the judg- ment. The protocol also records the source search, miss- ing evidence, included instruments and cutoff date. The result is a traceable comparison of policy content that keeps category differences visible. It evaluates what a portfolio provides on paper. It does not estimate enforce- ment success or policy outcomes. Figure 1: AI Policy Evaluation Framework 1∗ Corresponding author: kprab@uic.edu 1 Introduction Artificial intelligence is changing how governments carry out ordinary work. Agencies already use systems for language translation and image recognition (Mo- hamed et al. 2024; Li 2022). Other systems help detect fraud, process administrative data and allocate public resources (Obeng et al. 2024; Pi 2021; Rahman et al. 2024). AI is also used in public services and in deci- sions that affect citizens directly (van Noordt and Mis- uraca 2022; Engin and Treleaven 2019). In healthcare it can support diagnosis and drug development (Zeb et al. 2024). In hiring it can screen applicants or assist workforce management (Cai, Zhang, and Zhang 2024). Transport systems use it for planning and increasingly for automated operation (Jevinger, Zhao, and Persson 2024). These uses place AI inside institutions whose mistakes can affect access to services, employment, mo- bility and individual rights. Governments pursue AI because it can reduce rou- tine administrative work and help agencies make use of large datasets. It can support policy analysis, improve the delivery of some services and help allocate lim- ited staff or public resources (van Noordt and Misuraca 2022; Pi 2021; Rahman et al. 2024). Natural-language systems can make information easier to access, while forecasting and analytical tools can help officials exam- ine problems that would otherwise be difficult to study at scale (Mohamed et al. 2024). These are reasons to consider adoption. They are not results that follow au- tomatically from deploying a model. The value of a sys- tem depends on the quality of its data, its performance in the intended setting and the process around its use. Officials must also be able to review its outputs and cor- rect a decision when the system is wrong. Economic conditions add pressure to adopt these technologies. Global corporate investment in AI reached $252.3 billion in 2024. US private investment reached $109.1 billion, compared with $9.3 billion in China and $4.5 billion in the United Kingdom. Over the same period, the cost of querying a system at a fixed level of performance fell more than 280-fold between late 2022 and late 2024 (Maslej et al. 2025a). Lower costs expand the number of organizations able to use capable models. They also shorten the time available arXiv:2608.15417v1 [cs.CY] 15 Aug 2026 for governments to develop procurement rules and over- sight arrangements. The technical frontier has changed as quickly as the economics. DeepSeek released V3 in December 2024 and R1 in January 2025. It reported 2.788 million H800 GPU hours for V3 pretraining, and R1 showed how post-training could produce strong reasoning perfor- mance (DeepSeek-AI 2024; DeepSeek 2025). By mid- 2026, independent assessments placed leading Chi- nese systems about three to eight months behind the strongest closed US models. Frontier development in China had also broadened beyond one laboratory to sys- tems from Moonshot AI, Zhipu AI and Alibaba (Center for AI Standards and Innovation and National Institute of Standards and Technology 2026; Center for Strategic and International Studies 2026). Agentic systems add a different problem. A model can act through tools, com- plete several steps and change an external system be- fore a person reviews the result. Existing policy instru- ments often do not address the resulting questions about authority, monitoring and intervention directly (Staufer et al. 2026). The same systems create familiar public-law prob- lems in a new form. A model trained on historical ad- ministrative records can reproduce discrimination or di- rect errors toward groups that already face barriers to public services. A system that uses health or identity records must protect the information it receives. Fi- nancial and employment records require the same care. When a model contributes to a consequential decision, the responsible institution must specify who reviews the output and who can override it. The affected person also needs a clear explanation and a way to challenge the re- sult (Lim 2024; de Fine Licht and de Fine Licht 2020). Without these arrangements, an efficiency measure can weaken accountability instead of improving public ad- ministration. Implementation poses practical problems even when the policy objective is clear. Agencies need staff who can procure and test an AI system before use. They need people who can monitor it after deployment and decide when it should be changed or retired. Reliable data prac- tices and suitable technical infrastructure are also nec- essary. Funding must continue after the initial strategy is announced, and regulators with overlapping mandates need a workable way to coordinate. Workforce planning matters beyond technical teams because AI can change the work performed by public employees and by people in regulated industries (OECD 2024). Traditional rule- making and procurement often move more slowly than the technology they are expected to govern. Governments have responded with different policy instruments. Public investment and public-private part- nerships can fund infrastructure or research that no sin- gle institution would undertake alone (Mikhaylov, Es- teve, and Campion 2018). Regulatory sandboxes allow a system to be tested in a controlled setting before wider deployment (Truby et al. 2022). Technical standards and risk-management frameworks can state what evi- dence an organization should produce and how it should manage known risks (Cihon 2019; National Institute of Standards and Technology (NIST) 2023). Procurement rules can make those expectations a condition of selling to the state. Tax policy, grants and training programs can encourage adoption or direct development toward public needs (Fischer et al. 2021). These instruments cannot be treated as interchange- able. A statute can create a duty but leave its operation to later rules. A strategy can allocate resources without giving an affected person an enforceable right. Volun- tary guidance can explain a practice in more detail than binding legislation, yet create no penalty when an orga- nization ignores it. Institutional mandates and budgets determine whether any of these commitments can be carried out. National AI governance is therefore better understood as a policy portfolio than as one prominent law or strategy. Comparing only the best-known doc- ument can hide both the strengths and the gaps in the wider approach. International coordination has not produced a sin- gle model. The Bletchley Declaration focused attention on frontier-system risk. The Seoul process added vol- untary company commitments and cooperation among public institutions (UK Department for Science, Inno- vation and Technology 2023a; UK Department for Sci- ence, Innovation and Technology and Government of the Republic of Korea 2024). The United States and the United Kingdom did not sign the declaration issued at the 2025 Paris AI Action Summit (TechCrunch and Wiggers 2025). These events do not divide national ap- proaches into two simple camps. They show that a com- parative method must allow for different combinations of legal duties, voluntary practice and development pol- icy. The following sections develop such a method. Sec- tion 2 first examines the technical and institutional set- ting of eight focal jurisdictions, then compares their pol- icy portfolios with the units used by existing legal stud- ies and readiness indices. Section 3 defines a versioned national policy portfolio and explains how its provisions are scored. The resulting framework is designed to show what a policy provides on paper, how clearly it provides it and which evidence supports the judgment. It does not treat a documented commitment as proof of compliance or social benefit. 2 Background and Related Work 2.1 Cross-National Comparison of AI Capabilities Metrics for Comparison. National capability affects the resources available for AI policy and the setting in which that policy operates. A government with substan- tial public compute can support independent testing or domestic research directly. A government that depends on imported hardware and foreign cloud providers faces different constraints. Domestic semiconductor firms and an experienced technical workforce can also change Figure 2: Components of a national AI ecosystem what the state is able to build, inspect or regulate. None of these conditions determines the quality of regulation. They show which commitments are feasible and where external dependence can hinder implementation. The focal jurisdictions are China, India, Japan, Singa- pore, South Korea, the United Kingdom and the United States. The European Union is included as a suprana- tional comparator. Together they provide marked dif- ferences in technical capacity and economic conditions. They also differ in talent and regulatory approach (Gil and Perrault 2025; Dutta and Lanvin 2022). Figure 2 or- ganizes this context around compute infrastructure, the economic and political environment and domestic de- mand for AI expertise. These dimensions do not form a second index and do not contribute to the policy score. They help explain why the same policy commitment can be easier to carry out in one jurisdiction than an- other. Infrastructure for AI Compute. Computing capac- ity is divided into three forms because each provides a different kind of access. A national supercomputer is not equivalent to a private training cluster. Neither is equivalent to a commercial cloud region. • Scientific Supercomputing Facilities. Public agen- cies, universities and national laboratories operate high-performance computing centers. These facili- ties can make expensive scientific workloads avail- able to researchers or public bodies that could not fund them alone. Peak performance and accelerator counts describe the machine. Access rules show who can use it. A place in a published ranking does not establish that the system is suitable for model train- ing or open to outside researchers. • Private Company Clusters. Companies operate these facilities for model development and commercial services. A large cluster can support frontier re- search and deployment at scale. Its location can also affect domestic expertise and energy demand. Pub- lic sources rarely disclose a complete inventory or show how fully the hardware is used. The compari- son therefore reports documented facilities and com- mitments rather than presenting them as national to- tals. • Cloud Compute Regions. A cloud region is a local group of data centers through which a provider of- fers computing services. Local service can reduce la- tency and help an organization meet a data-residency rule. It also lets smaller users obtain computing re- sources without building a facility. The presence of a region does not show the price or scale of available AI accelerators. A general cloud region should not be treated as a large accelerator cluster. Economic and Political Environment. Economic capacity and external relations shape national AI de- velopment. The comparison records these conditions as context. It does not assume that a large economy has better policy or reduce political conditions to a single stability rank. • GDP and Economic Growth. GDP shows the scale of an economy, while growth records its recent di- rection. A larger fiscal and industrial base can sup- port public infrastructure or research funding. It can also fund regulatory staff and transition assistance. These measures do not show how much a govern- ment spends on AI or whether that spending pro- duces sound policy. • Political and External Context. Trade policy and export controls can determine whether firms can obtain advanced semiconductors. International rela- tions can also affect foreign investment and research partnerships. The field reports these concrete con- straints together with relevant national policy. The reviewed sources do not support one comparable sta- bility measure across all jurisdictions. • Regulatory Frameworks. AI-specific instruments of- ten operate beside general law and sectoral regula- tion. Executive policy can add standards or voluntary guidance. Legal force is recorded separately because a detailed recommendation and an enforceable duty can describe the same conduct while creating differ- ent obligations. • Resource Availability. AI development depends on electricity and semiconductors. It also requires suit- able data centers and network infrastructure. These physical inputs affect cost and scale. Their presence does not prove practical access because prices, trade restrictions or competing demand can still limit use. Demand and Talent in AI. Domestic demand affects which systems are developed and where pressure for adoption arises. Research institutions and the available workforce determine whether those systems can be built and tested locally. They also affect governance because the state needs its own technical and domain expertise. • Local Demand for AI Solutions. Adoption and pro- curement show where organizations are already us- ing AI. Investment provides another signal of ex- pected demand. The field considers major private sectors as well as public administration. These sig- nals can reveal pressure for clear rules, but they do not form a comparable market-size estimate unless the sources use the same definitions and period. • Research Institutions. Universities and public labo- ratories can train specialists or provide shared infras- tructure. Research networks can also support testing and independent expertise. Listing these institutions does not by itself measure the quality of their re- search or public access to it. • Talent Pool. Workforce estimates and education pro- grams indicate the available supply of AI exper- tise. Migration and reported shortages show where that supply is under pressure. The relevant pool ex- tends beyond researchers and engineers. Officials, lawyers, auditors and domain specialists are also needed to govern deployed systems. Sources define AI occupations differently, so their estimates are not combined into one ranking. Analysis of Selected Jurisdictions. The broader in- ventory in Appendix A covers thirteen jurisdictions. Eight were selected for the policy comparison. The se- lection was purposive because the framework needed to be examined against different levels of technical capac- ity and different forms of governance. The United States and China have large research and commercial ecosys- tems, but organize regulatory authority differently. The European Union provides a supranational legal model built around risk classification. The United Kingdom re- lies on existing sectoral regulators. Japan and Singapore make extensive use of coordination and industry-facing guidance. South Korea combines development policy with a comprehensive AI statute. India joins rapid ca- pability development and digital public infrastructure with data protection law and an evolving AI governance structure. This set is not a representative sample of gov- ernments and cannot establish global prevalence. Only these eight jurisdictions form the policy com- parison. Tables 1 and 2 summarize the context for that comparison. Policy status claims are current through August 15, 2026. Historical datasets retain their stated reference years. The tables deliberately keep policy context separate from the later score. A jurisdiction is not rewarded in the policy framework merely because it has more compute, a larger economy, or a larger AI workforce. 2.2 Cross-National Discussion of AI Policy Approaches The eight portfolios differ in legal force and institu- tional structure. They also place different weight on development policy and protective regulation. For this reason, the comparison evaluates a portfolio instead of searching for one national AI law. The discussion below identifies the differences that shape the framework. Ap- pendix B provides the fuller instrument-by-instrument comparison. European Union. The AI Act entered into force on August 1, 2024 (European Union 2021b; European Commission 2024b). It prohibits specified practices and places obligations on high-risk systems. Other provi- sions apply to systems subject to transparency duties and to general-purpose AI models. The duty depends on the system and on the actor’s role. A provider of a high- risk system must establish risk management and data- governance arrangements. It must also document the system, provide for human oversight and monitor per- formance after market entry. General-purpose AI provi- sions require documentation and information for down- stream providers. They also address copyright compli- ance, with additional duties for models that present sys- temic risk. Implementation is phased. Prohibited-practice and AI-literacy provisions applied from February 2025. Governance and general-purpose AI provisions applied from August 2025, alongside a voluntary Code of Prac- tice (European Commission 2024a, 2025c). Regulation 2026/1744 later revised the schedule for several high- risk obligations (Council of the European Union 2026; European Union 2026). The portfolio now extends be- yond the binding Act to standards and implementing work. Guidance and the Code of Practice are meant to help providers demonstrate compliance. Enforcement is divided between the AI Office, the Commission and in- stitutions at member-state level. The model gives rights and risk controls a clear legal basis, but its operation de- pends on coordination and adequate institutional capac- ity (European Parliamentary Research Service 2022). It also depends on the completion of technical standards. Legal force must therefore be assessed separately from clarity, resources and implementation. United States. The United States has no compre- hensive federal AI statute. Executive policy and ex- isting agency authority carry much of the federal ap- proach. Technical standards, federal procurement and research programs add further instruments, while state law supplies many binding rules for specific uses. Un- der the previous administration, Executive Order 14110 directed federal action on AI safety and security. It also addressed civil rights, procurement and govern- ment use. The Blueprint for an AI Bill of Rights stated nonbinding principles on safety and discrimination. Its other principles concerned privacy, notice and access to a human alternative (White House 2023, 2022). The federal direction changed in January 2025. Exec- utive Order 14110 was revoked. Executive Order 14179 then directed a new plan focused on US leadership and the removal of barriers to AI development (The White House 2025a,b). The July 2025 AI Action Plan contains more than 90 actions. They address innovation and in- frastructure as well as government adoption and inter- national policy (The White House / Office of Science and Technology Policy 2025). NIST continues to main- tain the voluntary AI Risk Management Framework, while sectoral agencies use powers derived from exist- ing law. States remain an important source of binding rules in particular domains (National Institute of Stan- dards and Technology (NIST) 2023; StateScoop 2025). Access to advanced chips has changed as well. The AI CountryInfrastructure for AI ComputeEconomic and Political Environ- ment Demand and Talent in AI SingaporeThe government committed S$270 million to expand the National Su- percomputing Centre and related training. Commercial providers also operate accelerator capacity and lo- cal cloud regions (Keat 2024; G 2025). The economy grew 4.8% in 2025. Budget measures committed S$1 billion over five years to AI and S$150 million to the Enterprise Compute Initiative. These invest- ments operate alongside nonbind- ing governance tools (Ministry of Trade and Industry, Singapore 2026; The Edge Singapore 2024; Mother- ship 2025; Personal Data Protection Commission Singapore 2020). The national strategy identifies fi- nance, healthcare, logistics, and public services as areas for wider use. NUS, NTU, and public re- search bodies support research and training. The strategy targets 15,000 AI practitioners by 2028, and Sin- gapore issued an agentic AI frame- work in January 2026 (Smart Nation Singapore 2019; Infocomm Media Development Authority, Singapore 2026). JapanABCI 3.0 became publicly avail- able in January 2025 with 6,128 NVIDIA H200 GPUs and a stated peak of 6.22 EFLOPS at FP16. Fu- gaku and private or cloud capacity extend the compute base (NVIDIA 2025a; Takano et al. 2024). A large economy and sustained technologyinvestmentsupport public and private AI programs. Human-centeredprinciplesand sectoral guidance now operate alongside the 2025 AI Promotion Act, which relies on guidance, information requests, and public disclosure rather than private-sector penalties(Wolf2025;Cabinet Office, Government of Japan 2025; Government of Japan 2025). Manufacturing,healthcare,and robotics create demand. The Uni- versity of Tokyo, AIST, and other institutions use shared infrastruc- ture for research, while government and university programs support AI education and training (Insights 2023; Harris 2024). IndiaThe PARAM series and National Supercomputing Mission provide public high-performance comput- ing. The IndiaAI compute portal re- ported more than 38,000 GPUs on- boarded by March 2026, alongside commercial cloud regions (Depart- ment of Science and Technology 2022; Digital India 2026; APAC Media 2023). Nominal GDP was estimated at about $4.15 trillion in 2025, with 6.6% real growth projected for fiscal year 2025–26. The IndiaAI Mission operates alongside the 2023 data- protection statute, 2025 rules, and voluntary AI Governance Guide- lines (International Monetary Fund 2025; Press Information Bureau, Government of India 2025; Min- istry of Electronics and Information Technology, Government of India 2025). Demand spans healthcare, agricul- ture, education, and urban services. IITs, IIITs, IISc, public agencies, and research hubs support devel- opment and training, while sources continue to report unmet demand for advanced skills (Singh et al. 2025; Peterson 2024; Arnab Kumar and Mahindru 2024; Stanly 2024). United King- dom ARCHER2, CSD3, and Isambard- AI provide scientific and AI com- puting capacity. London, Cam- bridge, Oxford, Manchester, and Edinburgh host additional commer- cial and research clusters (Research and Innovation 2021; Shainer 2021; University of Bristol 2025; Korolov 2025). Nominal GDP reached about £3.03 trillion in 2025. The government adopted all 50 recommendations of the AI Opportunities Action Plan and retained a regulator-led ap- proach built around nonstatutory principles, the AI Security Institute, and regulatory sandboxes (House of Commons Library 2026; Clif- ford and Department for Science, Innovation and Technology 2025; Department for Science, Innovation and Technology 2025; Rough and House of Commons Library 2026). Oxford, Cambridge, Imperial Col- lege London, University College London, and the Alan Turing Insti- tute support research and govern- ment collaboration. UKRI doctoral funding, industry partnerships, and international recruitment form part of the training system (Benn 2025; Sharma 2024). Table 1: Cross-National AI Ecosystem Comparison CountryInfrastructure for AI ComputeEconomic and Political Environ- ment Demand and Talent in AI United StatesEl Capitan, Frontier, and Aurora ranked second through fourth on the June 2026 TOP500 list. Commer- cial capacity includes large GPU clusters and custom accelerators from Google, AWS, and Cerebras. The National AI Research Resource pilot provides shared resources for academic and public-interest re- search (TOP500.org 2026b; Miller and Gelles 2024; Google Cloud 2026; Amazon Web Services 2026; Cerebras Systems 2026). Nominal GDP was about $30.5– 30.8 trillion in 2025. Executive Or- der 14179 and the July 2025 AI Action Plan emphasize innovation, infrastructure, and national leader- ship, while export controls on ad- vanced chips changed several times during 2025 (U.S. Bureau of Eco- nomic Analysis 2026; The White House 2025b; The White House / Office of Science and Technology Policy 2025; Bureau of Industry and Security, U.S. Department of Com- merce 2025). MIT, Stanford, CMU, federally funded institutes, and industry lab- oratories support a large research base. The United States remains a major destination for researchers and engineers, while immigration policy affects recruitment and reten- tion (Maslej et al. 2025b; Oschinski et al. 2025). European Union EuroHPC coordinates shared pub- lic supercomputing systems across member states. The InvestAI initia- tive targets C200 billion in public and private AI investment, includ- ing support for large shared facili- ties (European Union 2021a; Euro- pean Commission 2025b,a). GDP wasC16.22 trillion in 2024. The AI Act entered into force in 2024, and Regulation 2026/1744 re- vised parts of its implementation schedule. Collective digital policy and member-state implementation make the EU a supranational com- parator rather than a single national environment (Eurostat 2026; Euro- pean Commission 2024b; European Union 2026; European Commission 2026). Universities, national laboratories, and the ELLIS network support re- search across member states. EU programs support adoption in man- ufacturing, finance, healthcare, and public services, while private AI investment remains below the US level (Eurostat 2025; Tambiama Madiega with Rafał Ilnicki; Graph- ics: Eulalia Claros 2024; Stanford Institute for Human-Centered Arti- ficial Intelligence 2026). South KoreaSamsung and SK Hynix supply memory used in AI systems. Naver, Kakao, domestic cloud providers, and a government-industry plan for more than 260,000 GPUs support public and private compute (Park 2024; Chang-won 2021; NVIDIA 2025b). Nominal GDP was about $1.86– 1.87 trillion in 2025. The AI Ba- sic Act took effect on January 22, 2026, and distinguishes high-impact AI by sector and consequence from high-performance AI defined partly through a 10 26 operation threshold (International Monetary Fund 2026; Republic of Korea 2025; Jon 2026). Healthcare, finance, manufacturing, urban planning, and public services create demand. KAIST, Seoul Na- tional University, and industrial lab- oratories support research, patents, education, and workforce develop- ment (McFaul et al. 2023). ChinaLineShine ranked first on the June 2026 TOP500 list at 2.198 exaFLOPS. Huawei’s Atlas 900 A3 SuperPoD connects up to 384 Ascendchips,whiledomestic cloud providers and the East Data West Computing program sup- port a distributed compute base (TOP500.org 2026a; Huawei 2025; Alibaba Cloud 2026; Tencent Cloud 2026; State Council of the People’s Republic of China 2024a). Final revised GDP for 2024 was 134.81 trillion yuan, with reported real growth of 5.0%. The August 2025 AI Plus initiative, synthetic- contentrules,andAI-specific amendments to the Cybersecurity Law supplement existing platform and data regulation (National Bu- reau of Statistics of China 2025; Zhou and Interesse 2025; State Council of the People’s Republic of China 2025; Cyberspace Ad- ministration of China and others 2025; National People’s Congress Standing Committee 2025). Universities, public institutes, and laboratories at DeepSeek, Moon- shot AI, Zhipu AI, Alibaba, and other firms support a large re- search base. China produced 23.2% of AI publications in 2023, while independent evaluations placed its leading systems months behind the strongest closed US systems in 2026 (Stanford Institute for Human-Centered Artificial Intelli- gence 2025; Center for AI Stan- dards and Innovation and National Institute of Standards and Technol- ogy 2026; Center for Strategic and International Studies 2026). Table 2: Cross-National AI Ecosystem Comparison continued Diffusion Rule was rescinded before it took effect, and controls concerning China changed again later in 2025 (Bureau of Industry and Security, U.S. Department of Commerce 2025; RFA 2025). A score based on one ex- ecutive order would miss this dispersion and the change over time. The unit must be a federal portfolio frozen at a stated cutoff. Subnational law should either be ex- cluded or analyzed under a separately declared scope. United Kingdom. The United Kingdom retains the regulator-led approach set out in its National AI Strat- egy and 2023 AI White Paper (UK Government 2021; UK Department for Science, Innovation and Technol- ogy 2023b). The White Paper asks existing regulators to interpret five cross-sector principles within their cur- rent mandates. Safety, security and robustness form the first principle. The next two concern transparency and explainability, then fairness. The last two address ac- countability and governance, followed by contestability and redress. The approach uses the domain knowledge of bodies such as the Information Commissioner’s Of- fice and the Financial Conduct Authority. The Compe- tition and Markets Authority has a separate role within the same model. No single general AI regulator was cre- ated. The government published the AI Opportunities Ac- tion Plan on January 13, 2025, and accepted all 50 rec- ommendations (Clifford and Department for Science, Innovation and Technology 2025). The plan focuses on the capacity needed for wider adoption. It addresses compute and data access, together with talent and the ability of the state to use AI. It does not create a com- prehensive AI statute. In February 2025, the AI Safety Institute became the AI Security Institute and focused its public mandate on serious security risks (Depart- ment for Science, Innovation and Technology 2025). Later policy continued to emphasize coordination and controlled experimentation (Rough and House of Com- mons Library 2026). A regulator-led model can adapt rules to sector-specific harms, but only if the regula- tors have suitable authority and resources. Consistency across their interpretations also matters. The portfolio therefore makes interagency coordination and the limits of existing mandates central to assessment. Singapore. Singapore uses nonbinding frameworks alongside sectoral law. The Model AI Governance Framework gives organizations practices for internal governance and human involvement. It also addresses operations management and communication. AI Verify provides a testing and reporting structure intended to make organizational claims easier to examine (Personal Data Protection Commission Singapore 2020). Separate frameworks address generative AI and, from January 2026, agentic AI (Infocomm Media Development Au- thority and AI Verify Foundation 2024; Infocomm Me- dia Development Authority, Singapore 2026). This approach can update more quickly than legis- lation and can give firms operational guidance before a consensus exists on binding rules. It also depends heavily on voluntary adoption and does not itself create penalties or a general right of redress. A useful evalu- ation must therefore distinguish substantive detail from legal force. Singapore can score highly for specifying a practice while receiving a different score on whether that practice is mandatory. Monitoring and enforcement remain separate questions. Japan. Japan combines the Society 5.0 strategy with human-centered principles and sectoral guidance. The 2025 AI Promotion Act adds a statutory basis to that approach (Cabinet Office, Government of Japan 2019; Japanese Ministry of Economy, Trade and Industry 2020; Cabinet Office, Government of Japan 2025). The broader strategy links AI development to demographic pressure and productivity. It also presents AI as a way to improve public services and meet social needs. Human- centered principles supply a normative basis, while ministries and sectoral bodies translate them into guid- ance for particular settings. The AI Promotion Act was promulgated on June 4, 2025, and fully entered into force on September 1, 2025. It authorizes national planning and research support. The government can request information, issue guid- ance and disclose noncompliance. The Act does not es- tablish private-sector fines or criminal penalties (Gov- ernment of Japan 2025). Japan therefore shows why statutory status and coercive enforcement are different features. Its portfolio also makes socioeconomic objec- tives relevant to the framework because development and social use are part of the policy’s stated purpose. India. India combines development programs with digital public infrastructure. Data protection law and voluntary AI guidance add protective elements. The 2018 National Strategy for Artificial Intelligence em- phasizes inclusive growth. Its priority areas include healthcare and agriculture, as well as education and mo- bility. It also addresses smart cities and infrastructure (NITI Aayog, Government of India 2018). The IndiaAI Mission funds compute and datasets. Other programs support skills, startups and domestic model develop- ment. The implementing rules for the Digital Personal Data Protection Act were notified in November 2025, with major duties scheduled to apply in phases (Government of India 2023a; Press Information Bureau, Government of India 2025). The November 2025 AI Governance Guidelines propose a light-touch architecture and new forms of institutional coordination. They also propose an incident database and an AI safety institute (Min- istry of Electronics and Information Technology, Gov- ernment of India 2025). The portfolio therefore sepa- rates binding personal-data duties from broader AI rec- ommendations and capacity-building programs. It also shows why implementation feasibility matters. A pro- vision can be well stated while still depending on insti- tutional authority and technical expertise. Infrastructure and continuing resources must be assessed separately. Republic of Korea. The Republic of Korea enacted the AI Basic Act in January 2025, and the Act took effect on January 22, 2026 (Republic of Korea 2025; Jon 2026). The statute combines industrial support and governance obligations. It distinguishes high-impact AI used in specified consequential domains from high- performance AI defined partly through a 10 26 operation threshold. These classifications are not interchangeable and trigger different provisions. The Act also imposes transparency duties for specified generative AI outputs and requires domestic representation for certain foreign providers. The Personal Information Protection Act supplies rights and duties involving personal data and automated decisions. Nonbinding national ethics standards add principles involving human dignity and public benefit. They also address appropriate use (Personal Informa- tion Protection Commission, Republic of Korea 2023; Ministry of Science and ICT, Republic of Korea 2020). Korea’s portfolio therefore contains a horizontal AI statute and general data-protection law. Development policy and soft-law principles add further provisions. Evaluating the statute alone would miss protections and implementation conditions found elsewhere. Collapsing high-impact and high-performance AI would also mis- state the Act’s scope. China. China governs public-facing AI services through several linked instruments. Industrial strategy operates alongside platform regulation. Content rules also draw on cybersecurity and data law. The 2022 deep-synthesis provisions and 2023 generative AI mea- sures impose duties on providers. They address training data and security assessment, together with user protec- tion and content governance. Labeling duties also apply (Cyberspace Administration of China 2022, 2023). The scope of the two instruments differs. Each score should therefore identify the service and the provision that sup- ports it rather than refer generally to Chinese AI regula- tion. Joint labeling measures took effect in September 2025 and require explicit and implicit markers for spec- ified synthetic content (Cyberspace Administration of China and others 2025). The State Council issued the AI Plus initiative in August 2025, and amendments to the Cybersecurity Law added AI-specific provisions from January 2026 (State Council of the People’s Republic of China 2025; National People’s Congress Standing Committee 2025). No standalone national AI law had been enacted by the cutoff. The resulting model can add targeted rules quickly and connect AI governance to established cybersecurity and data institutions. It can also make the boundaries of the overall regime harder to see because obligations are distributed across instru- ments and enforcement information is not equally pub- lic across sectors. The portfolio illustrates why source coverage and institutional responsibilities must accom- pany any score. Emerging Governance Challenges. Several devel- opments cut across the national differences. Interna- tional work on frontier AI risk advanced through the Bletchley Declaration and the Seoul Frontier AI Safety Commitments (UK Department for Science, Innovation and Technology 2023a; UK Department for Science, In- novation and Technology and Government of the Re- public of Korea 2024). The United States and United Kingdom then declined to sign the declaration issued at the 2025 Paris summit (TechCrunch and Wiggers 2025). National institutions also changed their mandates and terminology. These shifts show that international coor- dination can produce shared practices without produc- ing one stable national model. The framework therefore considers participation and interoperability as evidence of harmonization. It also considers practical coopera- tion, rather than treating agreement with one declaration as the only relevant measure. Governance has also moved toward general-purpose and advanced models. The EU created dedicated general-purpose AI obligations. The United States re- sponded through evaluation and procurement rather than one general model law. Infrastructure policy and export controls form another part of its response. The United Kingdom assigned advanced-system research and evaluation to a specialized institute. These instru- ments differ in legal form and purpose, but each re- sponds to capabilities that can affect many downstream sectors. Category 1 therefore asks whether the portfolio covers the relevant systems and risks. Credit does not depend on the use of a particular label such as frontier or general-purpose AI. Agentic systems create a newer coverage problem. Deployed agents can take multi-step actions through tools, yet public safety disclosure remains limited (Staufer et al. 2026). Singapore issued a dedicated gov- ernment framework for agentic AI in January 2026 (Infocomm Media Development Authority, Singapore 2026). Other portfolios can still govern an agent through general risk duties or sectoral law. Human-oversight and cybersecurity provisions may also apply. The ab- sence of the label therefore does not prove the ab- sence of coverage. The relevant question is whether the portfolio addresses the capability and the operations it permits. This distinction motivates separate criteria for advanced-system safety and lifecycle coverage. It also explains the need to ask whether a policy is reviewed as technical conditions change. 2.3 Related Work The preceding comparison shows why an evaluation method must identify what it is evaluating. Some stud- ies describe a national AI ecosystem. Others compare legal models or ask whether a government is ready to adopt AI. Work on responsible AI often includes so- cial conditions and observed implementation. Company rubrics assess organizational commitments. Each unit supports a different conclusion. A country can have strong technical capacity and a thin regulatory portfo- lio. A statute can state detailed duties while the insti- tution responsible for them lacks staff or authority. A readiness index can identify enabling conditions with- out showing which legal provision creates a particular obligation. Comparative legal studies explain how regulatory philosophies and instruments differ across the United States, the United Kingdom, the European Union and China (Weiyue Wu 2023; Chun, Schroeder de Witt, and Elkins 2024). One line of work uses an “art, craft and science” account to explain how policy develops. More recent comparisons examine the distinct regulatory phi- losophy of each jurisdiction. These studies clarify the institutional and political choices behind a legal model. Their usual purpose is to organize legal difference, not to provide an operational scoring protocol that another evaluator can apply to the same defined portfolio. They also show why legal design cannot be inferred from a country’s technological capacity. Meta-frameworks and research on actionable AI principles address a related problem (Almeida, dos San- tos Jr., and Farias 2020; Stix 2021). They connect ethi- cal commitments to regulatory process and institutional responsibility. They also explain why a principle needs implementation support and some means of adaptation. This work is mainly conceptual or procedural. It does not always state how an evaluator should score a pro- vision that appears in several instruments. Nor does it necessarily separate a detailed nonbinding recommen- dation from a less detailed legal duty. Those distinctions become unavoidable in a cross-national score. Several national assessments use a broader unit than the one adopted here. UNESCO’s Readiness Assess- ment Methodology examines law and social conditions through a national multi-stakeholder process (UNESCO 2023). Economic and educational conditions are in- cluded, together with scientific and technical capacity. Its purpose is to support national diagnosis and policy dialogue. That breadth matters because governance de- pends on more than formal policy. It also means that a UNESCO readiness assessment and a score based on policy text answer different questions. Large cross-national indices make this difference clearer. The 2026 Global Index on Responsible AI cov- ers 135 countries and jurisdictions. It uses 68,138 data points across 38 indicators and includes evidence about implementation and civil-society participation (Adams et al. 2026). The AGILE Index covers 40 countries through four pillars, 17 dimensions and 43 indicators. It reports constraints as well as scores (Zeng et al. 2025). The 2025 Government AI Readiness Index covers 195 governments. It asks whether a government can use AI for public benefit by examining government capac- ity, the technology sector and data infrastructure (Ox- ford Insights 2025). These projects show that structured cross-national evaluation is feasible. They also reveal differences in state capacity that a document review cannot capture. The present framework draws a narrower boundary. A readiness measure can give weight to the strength of the technology sector or the supply of skills. It can con- sider public-sector adoption and the availability of use- ful data. A responsible AI index can include observed implementation and civil-society evidence. The policy score developed here asks what a declared portfolio pro- vides and how specifically it provides it. Economic ca- pacity is reported as context, not converted into a higher policy score. A readiness ranking is not treated as evi- dence that an individual legal duty exists. The score also does not infer an outcome from the words of a policy. Corporate grading rubrics provide a closer method- ological analogy. They use tiered criteria to assess fron- tier safety frameworks and make the basis of a grade visible (Alaga, Schuett, and Anderljung 2024; Stelling et al. 2025). This allows a reader to distinguish a spe- cific commitment from a vague statement. Their unit, however, is a company’s voluntary safety framework. A national portfolio can include legislation and exec- utive policy. It can also include a strategy, an institu- tional mandate or a procurement rule. Guidance and a budget commitment can provide further evidence. The same practice can therefore appear as a legal duty in one jurisdiction and as a voluntary recommendation in another. Domain-specific frameworks provide much of the substantive material needed for a national rubric. Healthcare governance supplies detailed questions about high-stakes use (Reddy et al. 2021). Privacy and security frameworks clarify data and system safeguards (Olukoya 2022). Research on legal effectiveness iden- tifies features needed for a rule to operate (Zuiderwijk, Chen, and Salem 2021). Work on system evaluation and organizational readiness contributes further evidence questions (Xia et al. 2024; Holmstr ̈ om 2022). These frameworks often provide more depth within their do- main than a national comparison can. They do not at- tempt to connect that depth to the entire national portfo- lio, where infrastructure and enforcement matter along- side individual rights. Advanced-system risk and eco- nomic transition also need to remain visible. The unresolved problem is therefore methodological, not a lack of governance concepts. A national policy evaluation needs a declared unit so that the evaluator cannot choose only the most favorable document. Each score needs a citation to the provision that supports it. The method must separate the detail of a commitment from its legal status because those attributes can move independently. Category results are also necessary. Oth- erwise a high total can conceal a weak result for safety, rights or institutional capacity. Policy change creates another problem. A new law can replace an executive instrument. Later guidance can supply details that were absent when the law was adopted. A budget can fund an institution only after its mandate has been announced. Court and regulatory de- cisions can change the practical scope of the same text. A score without a cutoff can therefore combine instru- ments that were never in force at the same time. Missing evidence must also be handled directly. Gov- ernments publish official material in different languages and with different levels of detail. An unsuccessful search is not always evidence that a provision is absent. Treating it as absence would penalize a portfolio be- cause it is harder to access, even when the underlying policy is unknown. The method therefore records the source search and creates a separate not-scorable state. It also reports how much of the rubric could be scored. National AI policy crosses several fields because its problems do. A principles-based review can miss en- forcement and resources. A technical-risk review can miss individual rights or distributional effects. An in- novation index can overlook the controls needed for high-consequence systems. The framework in Section 3 brings these questions into one instrument but reports their results separately. Appendix C records the main sources used in the design synthesis and explains how their units differ from the one used here. 3 Evaluation Framework 3.1 Framework Development and Design The framework was developed through a purposive de- sign synthesis rather than a systematic review. That choice follows from the research aim. The purpose is to construct and explain an evaluation instrument, not to estimate how often a concept appears in a fixed body of literature. Technical AI safety research was used to identify problems of misuse and control. It also in- formed the treatment of testing, intervention and longer- term resilience (Hendrycks, Mazeika, and Woodside 2023; Slattery et al. 2025; Ji et al. 2025). Existing gov- ernance and readiness frameworks supplied concepts concerning rights and institutions. They also informed the treatment of enforcement, participation and pol- icy change. Public instruments from the eight focal ju- risdictions showed how these concepts appear in ac- tual laws and strategies. Standards, budgets and official guidance provided further examples. The synthesis then translated each concern into a question that could be answered from public policy ev- idence. A general concern about model misuse, for ex- ample, is too broad to score. It becomes a question about what testing is required and who is responsible for it. Separate questions ask whether the policy controls access and requires incident reporting. Enforcement is examined on its own. Accountability is treated in the same way. Responsibility allocation is distinct from in- dependent oversight, while liability and redress require their own evidence. This translation is necessary be- cause naming a risk does not show that a government has created an operating response. Candidate concepts were grouped by the policy prob- lem they addressed, then checked against the eight port- folios. An item was divided when it joined features that can differ in practice. The detail of a rule, for example, can be strong even when its legal force is weak. Items were merged when they required the same evidence or would count one provision twice. A sub-criterion was retained only when it named a distinct feature that an evaluator could identify and justify from the declared sources. Appendix C records the principal frameworks used in this process. The design gives sustained attention to serious AI risk without treating safety as the whole of national pol- icy. Category 1 examines advanced systems and mis- use. It also addresses testing, human intervention and the resilience of supporting infrastructure. The remain- ing categories ask whether a policy can be implemented and whether its scope is adequate. They preserve indi- vidual rights and the distributional effects of AI as sep- arate concerns. Development policy and responsible in- novation therefore remain visible without being allowed to compensate silently for weak safety or rights provi- sions. The scoring system supports comparison and diagno- sis. It is not intended to create a false sense of precision. Ordered anchors distinguish an absent provision from a general statement. They then distinguish a partial mech- anism from a clear operating provision and from a fully specified response. Category and criterion results carry more information than the overall mean. The number remains useful because it forces the evaluator to state which evidence changes the judgment. It also permits the same portfolio to be scored again under the same rule. Every score must be traceable to a source, and every assessment must state when the portfolio was frozen. A later implementing rule can change the meaning of an earlier result. The creation of an institution or a change in its mandate can do the same. Withdrawal of an instru- ment also matters. Each such change produces a new portfolio version. The older result remains interpretable because its sources and cutoff are preserved. The unit of analysis is therefore a versioned na- tional policy portfolio. It includes official instruments that were in force or had been adopted by the cutoff. Depending on the declared scope, the instruments can be laws or regulations. Strategies and executive instru- ments can also form part of the portfolio, as can of- ficial implementation guidance. Institutional mandates and budget commitments are included when they sup- ply evidence for a criterion. The score sheet lists every instrument and gives its date or version. Subnational in- struments are excluded unless the assessment declares a different scope. The European Union is treated as a supranational comparator. Capability measures in Sec- tion 2.1 provide context but do not affect the score. The protocol requires evidence for each judgment and keeps legal force separate from substantive detail. It reports category results so that a single total does not hide where a portfolio is strong or weak. It also supports rescoring when the portfolio changes. These choices make an assessment auditable. They do not es- tablish that the rubric is reliable or valid in every legal and administrative setting. The instrument must still be tested by multiple evaluators. Its sensitivity to weight- ing should also be examined, and later work must com- pare the score with evidence about implementation. 3.2 Framework Structure Table 3 shows the framework’s five categories and 25 criteria. Appendix D supplies the sub-criteria and the evidence prompts beneath them. This hierarchy sepa- rates a broad policy objective from the provision an evaluator can locate in a portfolio. A category names a main dimension of governance. Its criteria divide that dimension into distinct questions, while the sub-criteria provide the unit of scoring. The listed indicators di- rect the evidence search. They do not receive separate scores. The default scheme gives equal weight to each category. Section 3.3 explains the remaining weights. AI Risk Governance and Safety. This category eval- uates provisions for high-consequence risks from cur- rent and advanced AI systems. Safety standards ask whether the portfolio identifies the systems and risks subject to stronger control. The standards should also explain what testing is required and connect that re- quirement to an accepted method or responsible institu- tion. Red teaming and auditing examine evaluation be- fore and after deployment. The rubric considers whether reviewers have enough independence and access to do the work, then asks whether an adverse finding leads to remediation. Human oversight concerns both authority and technical means. A responsible person must be able to pause or restrict a system, correct its operation or re- tire it when continued use creates unacceptable risk. The remaining criteria address the boundaries and longer-term conditions of safe operation. System inter- facing covers access control and containment. It also considers connected services, tool use and the security of the surrounding supply chain. Long-term governance examines misuse and concentration. Infrastructure re- silience asks whether critical services can continue and whether an institution can track risks that persist be- yond one deployment cycle. Sub-criterion 1.2 covers advanced or general-purpose systems regardless of the terminology used. A policy can receive credit without using terms such as AGI or frontier AI when its oper- ative provisions cover the relevant systems and risks. Criterion 5 concerns resilience over longer horizons. Criterion 15 asks a different question about monitoring technical change and updating the policy. Keeping them separate reduces double counting. Effectiveness and Feasibility. This category assesses the provisions that make implementation possible. A policy can state a strong objective while leaving no in- stitution responsible for carrying it out. Institutional ca- pacity therefore begins with a clear mandate and suit- able authority. The responsible body also needs staff, expertise and stable funding. Its independence and abil- ity to coordinate with other bodies are assessed sep- arately. Clarity asks whether regulated actors and af- fected people can determine the scope of a rule. They should be able to identify the duty and any exception. Enforceability examines the rule’s legal status and the means by which a violation can be detected. It also con- siders corrective action, sanctions and routes for review. Measurability asks whether implementation can be tracked. A measurable policy states its objective and identifies a suitable baseline. It sets an indicator and re- porting period, then assigns a body to collect or pub- lish the result. Feasibility also depends on resources. The policy should account for administrative cost and technical infrastructure, as well as workforce needs and burdens placed on regulated actors. Incentive alignment asks whether tools such as procurement or grants re- ward compliance and responsible development. Liabil- ity and market access can serve a similar function. The word effectiveness refers here to design and implemen- tation readiness before an outcome is observed. A score in this category is not evidence that the policy caused an outcome. Comprehensiveness and Scope. This category eval- uates whether the portfolio leaves a major part of AI development or use outside its coverage without ex- planation. Lifecycle coverage begins with research and design. It continues through data collection and train- ing, then examines evaluation and deployment. Moni- toring after deployment matters as much as incident re- sponse and later modification. The final stage is decom- missioning. No single instrument must cover the entire lifecycle, but the portfolio should allocate responsibility across it. The range-of-risks criterion begins with technical failure and deliberate misuse. It then considers discrim- ination and privacy loss. Security, labor effects and en- vironmental cost require separate attention because they arise through different mechanisms. Stakeholder inclu- sion asks who can contribute while a policy is made and put into effect. It also considers later review and access to redress, with particular attention to affected groups and expertise outside government or industry. Inter- national harmonization concerns standards and cross- border enforcement where national action is insuffi- cient. Research cooperation and technical interoperabil- ity can support the same aim. Adaptability asks whether the portfolio monitors technical change and reviews its assumptions. Credit depends on a defined update pro- cess, not the use of any particular technical label. User Rights, Protection and Agency. This category assesses protections for people affected by AI systems and by decisions made with them. Fairness and non- discrimination require a way to assess unequal treat- ment. The portfolio should also address mitigation and continuing monitoring, then provide a remedy when harm occurs. Transparency is not the same as explain- ability. A public description of a system serves a dif- ferent purpose from the information needed to under- stand a particular decision. The affected person and the regulator often need different information. Privacy and data protection begin with lawful collection and limits on purpose. Security and controls on later sharing are also relevant. Access, correction and deletion rights de- termine what the individual can do after data have been collected. Category 1. AI Risk Governance and Safety 1. Safety Standards2. Red Teaming and Auditing3. Human Oversight and Interventions 4. System Interfacing and Delimitation5. Long-Term AI Governance and Infrastructure Resilience Category 2. Effectiveness and Feasibility 6. Institutional Capacity and Authority7. Clarity and Specificity8. Enforceability 9. Measurability10. Resource Requirements11. Incentive Alignment Category 3. Comprehensiveness and Scope 12. Coverage of AI Lifecycle13. Range of Risks Addressed14. Stakeholder Inclusion 15. Adaptability to Technological Advancements 16. International Harmonization Category 4. User Rights, Protection and Agency 17. Fairness & Non-discrimination18. Transparency & Explainability19. Privacy & Data Protection 20. Accountability & Oversight Category 5. Socioeconomic Impact and Innovation 21. Societal Benefit22. Innovation23. Economic Growth 24. Social Equity25. Public Trust Table 3: Overview of AI Policy Evaluation Framework Criteria Accountability asks who is responsible when work is divided between a developer and a deployer. Vendors and public bodies can have separate duties, while an in- dividual official can retain authority over the decision. The portfolio should also provide independent oversight and a route for complaint or appeal. Liability and mean- ingful redress are examined rather than assumed. These protections overlap in practice but are scored separately. A portfolio can provide notice without an explanation. It can grant data rights without allowing a person to chal- lenge an automated decision. It can also name a respon- sible institution without providing an effective remedy. Socioeconomic Impact and Innovation. This cate- gory examines what the portfolio seeks to enable and who is expected to benefit. Societal benefit asks whether a priority use responds to an identified public need. It then considers access and evaluation, together with the mechanism by which public value is expected to arise. Innovation policy can support research or shared in- frastructure. Economic policy can support startups and technology transfer, while competition policy affects who can enter the market. These criteria do not re- ward growth language alone. A higher anchor requires a program and a responsible institution, supported by resources or comparable implementation detail. Workforce measures include training and support for people whose work changes. Labor protection is a sep- arate concern, as is the ability of a public institution to hire and retain expertise. Social equity asks whether access is distributed across regions and groups. It also considers whether vulnerable groups bear a greater share of the risk. Public trust is assessed through con- crete measures such as communication and consul- tation. Transparency and credible assurance can con- tribute, but a general promise that the public will ac- cept AI does not. The category scores commitments and implementation provisions documented in the portfolio. It does not score later economic growth or changes in public confidence. Distributional outcomes also require separate evidence. 3.3 Scoring Methodology and Aggregation. An evaluator assigns one score to each sub-criterion. The listed indicators are evidence prompts for that judg- ment. They are not separate items that receive separate scores. Each score sheet records the cited provision, a short rationale, and any source-search notes. Before scoring begins, the portfolio and the search protocol are frozen. Official consolidated text is pre- ferred. Implementing guidance is reviewed in the orig- inal language or through a documented translation. A secondary source can help locate an instrument, but an available primary source should support the score. The record identifies every document searched and the date of the search. It also states the terms used to locate rel- evant provisions. The process begins with an inventory rather than a score. The evaluator first states the jurisdictional level and cutoff. The declared scope explains which types of instrument are included and how subnational law is treated. It also gives the language procedure. Each in- strument is then recorded under its official title with its date and legal status. A source link or official publica- tion reference completes the entry. This inventory pre- vents an assessment from combining superseded and current provisions. It also prevents an evaluator from adding a favorable instrument only after a weak result becomes visible. The portfolio is then reviewed against every sub- criterion. A keyword search can locate a candidate pro- vision, but the surrounding text determines whether it applies and which actor it binds. Later instruments must be checked for amendment or replacement. The score cites the operative text and explains why it satisfies the selected anchor. When several instruments jointly sup- port the score, the rationale explains what each one con- tributes. This evidence trail does not remove judgment. It makes the judgment reviewable. Sub-criterion Assessment on a 1–5 Scale. The five anchors are ordered descriptions. The numbers support aggregation, but they do not establish equal intervals be- tween adjacent anchors. 1. Absent. A defined search of the included sources finds no relevant provision. 2. Minimal/Vague. The portfolio acknowledges the is- sue but gives no specific commitment or responsible actor. It provides little operating detail. 3. Basic/Partial. The portfolio includes a concrete pro- vision, but its scope or authority is incomplete. Im- portant operating detail is missing. 4. Substantial/Clear. The portfolio includes clear pro- visions that cover the main evidence prompts and identify how the commitment operates. 5. Comprehensive/Specified. The portfolio covers the evidence prompts in detail. It specifies the respon- sible actor and procedure, together with scope and review where relevant. Enforcement is stated when the sub-criterion requires it. The evaluator selects the highest anchor whose de- scription is supported as a whole. One indicator cannot by itself justify a 5, and the indicators are not averaged as independent questions. The term comprehensive de- scribes coverage and specification under the particular sub-criterion. It does not mean that the policy is univer- sally optimal or that implementation will succeed. If the declared search does not allow the evaluator to determine whether a provision exists, the item is marked not scorable. It is not assigned a 1. The item is ex- cluded from its parent mean and the remaining child weights are renormalized. The result states the share of sub-criteria that received a score. A score of 1 is used only when the declared source search supports a find- ing of absence. If every child of a criterion or category is not scorable, that parent is also not scorable. The analysis does not report an overall score when an entire category is not scorable. This distinction matters when official information is incomplete. It also matters when a document is un- available in a language that can be assessed reliably or when delegated guidance cannot be located. Absence is a finding about the declared source set. Not scorable is a finding about the limits of the assessment. The cover- age rate and search notes keep missing evidence from disappearing inside an apparently precise total. Weighted Aggregation Let s cjk denote the score for sub-criterion k under criterion j in category c. The over- all score is S = 5 X c=1 w c X j∈J c w j|c X k∈K cj w k|cj s cjk .(1) Weights sum to one at each parent. The default gives every category a weight of 20%. Criteria receive equal weight within their category, and sub-criteria receive equal weight within their criterion. Category 2 has six criteria. Its table displays each weight as 16.67%, but the calculation uses the exact value 1/6. These defaults replace the unspecified weights in Appendix D. An analysis can use custom weights only when they are declared before scoring and justified with the re- search question. The default result must also be re- ported. Results produced under different weighting or omission rules are not directly comparable. Equal weighting is a transparent baseline, not a claim that every governance question has identical so- cial value. It prevents an evaluator from tuning the weights after seeing which jurisdiction benefits. A tar- geted study can assign different weights, but it should also show the result under the default. Sensitivity anal- ysis is especially important when totals are close. It is also needed when missing evidence causes substantial renormalization. Final Score Interpretation (1-5 Scale) The output reports the overall score and the result for each cate- gory. Criterion scores provide a more detailed view. The record also states coverage and includes the instrument list with its cutoff. The total summarizes alignment with the rubric. It is not a causal estimate. A high result for innovation does not erase a low result for safety or rights, and the category scores keep that difference vis- ible. Comparisons should begin with the category results and evidence notes rather than the total alone. Two port- folios can reach the same overall score through differ- ent strengths. One can rely on legal force, while another provides more detailed safety provisions or develop- ment policy. Rights and institutional capacity can differ even when the totals match. A small numerical differ- ence should not be treated as meaningful until evalua- tor agreement is known and the result has been tested against alternative weights and missing-data rules. When two evaluators score the same portfolio, agree- ment should be computed before they discuss their dif- ferences. Weighted Cohen’s kappa is suitable for the ordered scale (Cohen 1968). The study must declare whether the calculation uses linear or quadratic weights. More than two evaluators can be assessed through an ordinal form of Krippendorff’s alpha (Krippendorff 2018). Disagreements can then be resolved, while the record retains the pre-consensus scores and the agreed rationale. Every update creates a new portfolio version rather than silently replacing the old score. Annual reassess- ment is a useful default, with an additional update af- ter a major legal or institutional change. This procedure permits comparison over time while preserving the evi- dence behind each score. The same protocol can support an expert panel or a commissioned audit. It can also be used in a public re- search project. In every case, evaluators should score in- dependently before discussion. Consensus can resolve interpretive differences for the final record. The pre- consensus scores remain necessary because they show reliability and identify anchors that require revision. 4 Discussion and Future Work Contribution. Existing work provides comparative le- gal analysis and measures national readiness. Gover- nance indices examine broad country conditions, while company rubrics grade voluntary commitments. De- tailed frameworks also exist for particular sectors. The remaining difficulty is to turn their useful concepts into a traceable comparison of national policy portfo- lios. The instruments in such a portfolio do not share the same legal status or scope. They also differ in the amount of operating detail they provide. The framework narrows the task to documented policy design and im- plementation readiness. It asks what the portfolio pro- vides and which source supports that finding. It then asks who is responsible and how the provision is meant to operate. The category structure preserves differences that a single list of principles can hide. A technical safety obli- gation is not equivalent to an institution with enforce- ment authority. An individual right serves another pur- pose, while an innovation program addresses the devel- opment side of the portfolio. Category 1 examines tech- nical risk governance and the resilience needed to sus- tain it. Category 2 asks whether the institutions named in the policy can act. It also examines authority and re- sources, together with the incentives and measurement needed for implementation. Category 3 concerns cover- age. It tests whether the portfolio follows AI across its lifecycle and responds when the technology changes. The remaining categories keep the social purpose of policy visible. Category 4 examines the protection and agency of people affected by AI. It requires more than a statement of ethical values because rights need a responsible institution and a route to redress. Cate- gory 5 considers public benefit and economic policy. It also asks how workforce change and unequal access are addressed. Reporting these categories separately pre- vents a strong development program from compensat- ing silently for weak rights or safety provisions. Narrower frameworks remain useful. A healthcare framework can examine the evidence needed for clin- ical use in greater detail than a national rubric (Reddy et al. 2021). Work on privacy or market regulation can do the same within its domain (Zuiderwijk, Chen, and Salem 2021). Those approaches cannot by themselves show whether the national portfolio creates a cross- sector oversight body or gives it adequate resources. They can also omit a general incident-reporting rule or a process for reviewing emerging risk. A broad gover- nance index has the opposite limitation. It can reveal na- tional conditions without identifying the provision that creates a right or duty. The present rubric connects these levels while keeping its claim limited to policy content. Legal force in context. Sub-criterion 8.4 illustrates why scope and legal status must be recorded. If the de- clared portfolios are limited to the named AI-specific instruments, the EU AI Act and the Republic of Ko- rea’s AI Basic Act are binding primary legislation. Each would receive a 5 for legal status and force. Singapore’s Model AI Governance Framework is detailed official guidance, but it is nonbinding. It would receive a 2 on this sub-criterion (European Union 2021b; Republic of Korea 2025; Personal Data Protection Commission Sin- gapore 2020). These scores answer one question about legal force. They do not compare substantive quality or practical influence. They say nothing about the enforce- ment record or the overall portfolio score. The example also shows why legal force cannot stand in for policy quality. Singaporean guidance can de- scribe an operating practice more clearly than a broadly worded statute. A binding law can create authority while postponing important obligations or leaving de- tails to later standards. A strategy can allocate sub- stantial resources without creating an individual right. Treating all official documents as equivalent would ig- nore these differences. Treating statutory force as a complete measure would make the opposite error. Using and interpreting the framework. A complete assessment begins by defining the portfolio. The record states the cutoff and jurisdictional level before any score is assigned. It identifies the source types and transla- tion procedure, then lists the included instruments. Each sub-criterion receives one score supported by cited ev- idence and a short explanation. Independent evaluators apply the same search and scoring rules before they dis- cuss a disagreement. The final record retains their initial scores and the agreement statistic. It also retains the re- solved score and explanation. Equal weighting provides a common starting point for a general comparison. A narrower research question can justify different weights. A study of AI in public administration, for example, can place greater weight on procurement and institutional authority. Rights and access to redress can also receive more weight in that setting. A study of advanced-model governance can in- stead emphasize testing and intervention. Misuse and system resilience would receive closer attention. The analysis should still report the default result and dis- close the custom weights. Otherwise a reader cannot tell whether a difference comes from the policy or from the priorities built into the calculation. Interpretation should begin with the category and cri- terion results. The overall score is convenient, but the supporting evidence lies below it. A low institutional- capacity result can explain why an enforcement pro- vision has no responsible body. Limited resources can also constrain monitoring or prevent a policy from be- ing updated. These relationships matter, but one weak provision should not be counted several times without distinct evidence. The score sheet should support each sub-criterion separately. The written analysis can then explain how the weaknesses interact. The framework distinguishes several kinds of in- completeness because they imply different responses. A score of 1 means that the declared search supports the absence of a provision. A score of 2 means that the portfolio acknowledges the issue without explain- ing how the response will operate. A not-scorable entry means that the evidence did not support either conclu- sion. An absent provision can require a new commit- ment. A vague provision can require implementing de- tail. A not-scorable item can require better publication or a reliable translation before any substantive judgment is justified. Detailed application will take time. A complete port- folio can spread one policy question across legislation and guidance issued by different bodies. Amendments and superseded versions must be checked before the op- erative text is identified. The evaluator then has to apply a large rubric without treating a keyword match as suf- ficient evidence. This cost limits rapid assessment, but much of it follows from the breadth of national AI gov- ernance rather than from the scoring scale itself. A pub- lic instrument inventory and reusable evidence record could reduce later work without replacing expert judg- ment. Limits of document review. A detailed rubric re- duces unstructured judgment but cannot eliminate inter- pretation. Evaluators can disagree about whether a pro- vision is sufficiently specific. They may also disagree about whether two instruments jointly satisfy one an- chor. A legal term can have a different practical mean- ing across jurisdictions even when the translation looks similar. Independent scoring and agreement statistics expose these disagreements. Legal and linguistic exper- tise is especially important when the rubric addresses fairness or accountability. Institutional independence also needs to be understood in its national setting. The framework favors features that can be docu- mented. An explicit duty is easier to score than an in- formal practice. A named institution and published pro- cedure leave a clearer evidence trail than coordination that occurs inside government. A stated budget or re- view period can be assessed in the same way. This pref- erence improves auditability, but it can underrate soft law that strongly shapes conduct. It can also underrate a public institution that works effectively without de- tailed published rules. The opposite error is possible as well. A carefully drafted law can receive a strong de- sign score even when it is weakly enforced. Effective- ness and feasibility are therefore limited to design and readiness. The category does not measure an outcome. Information availability creates a related limit. Gov- ernments publish official instruments in different forms and languages. Some implementation evidence remains inside agencies or regulated firms. A secondary source can reveal that a missing instrument exists, but using it in place of official text can introduce error. The search protocol and not-scorable state make this problem visi- ble. Coverage reporting shows how much of the rubric rests on evidence. None of these measures can create information that is not public. The rubric also contains normative choices. It gives weight to advanced-system safety and enforceable rights. Institutional capacity, public participation and equitable access also affect the score. Responsible in- novation remains part of the assessment. Another study could reasonably choose different weights or refine an anchor for a particular legal system. A revised rubric should be versioned and justified rather than presented as a neutral correction. It should also be tested across different administrative traditions before small score differences are treated as substantive rankings. Future work. The immediate empirical task is to ap- ply a frozen version of the rubric to complete portfolios using multiple evaluators. That study should publish the inventory of instruments and the source-search record. Evidence extracts and completed score sheets would allow another evaluator to inspect each judgment. Ini- tial scores should be preserved alongside the agreement statistic and the final consensus result. Sensitivity anal- ysis would show whether the conclusions depend on the default weights or missing-evidence rule. Together these materials would reveal which anchors work con- sistently and which require revision. A maintained public implementation could make later assessments easier to audit. It should preserve each version of the rubric and each dated national portfo- lio. Official sources could be linked to examples that show why a provision satisfies a particular anchor. A change log would explain later revisions. Expert review would remain necessary because a collaborative repos- itory cannot determine legal scope or factual accuracy by vote. Its main value would be traceability and reuse, not automated scoring. Specialized modules could add depth for foundation models or agentic systems. Public-sector procurement could use a module of its own. Healthcare and finance also raise domain questions that the national core can- not address fully. A module should not change the core score unless that change is declared in advance. Sta- bility in the core is needed for comparison over time, while the modules can respond to applications that re- quire more specific evidence. Outcome validation is a separate long-term task. Later studies can compare policy-design scores with enforcement actions and compliance records. Inci- dent data and institutional budgets can show whether the stated mechanisms operate. Public complaints and adoption data provide other forms of evidence, while social or economic outcomes answer still broader ques- tions. Such work could test whether the documented features measured here predict implementation or pub- lic benefit. It requires longitudinal evidence and a causal design that the present content-based framework does not provide. 5 Conclusion National AI governance cannot be understood through the most prominent law or strategy alone. A binding provision can depend on later guidance and on an in- stitution with enough authority to apply it. A develop- ment strategy can supply resources without creating an enforceable right. A detailed voluntary framework can shape practice while leaving compliance to the orga- nization. The comparison of eight jurisdictions shows these elements being combined in different ways. The European Union and the Republic of Korea place comprehensive AI statutes at the center of their portfo- lios. The United States distributes federal policy across executive action and existing agencies, while the United Kingdom asks sectoral regulators to apply common principles. Singapore and Japan rely more heavily on guidance and coordination, together with measures that promote adoption. India joins development policy and data protection with an emerging AI governance struc- ture. China’s approach is spread across rules for plat- forms and content, supported by cybersecurity and data law. These differences make the portfolio, rather than one document, the appropriate unit of analysis. The framework defines that unit as a national portfo- lio frozen at a stated cutoff. Its 25 criteria ask whether serious risks are governed and whether the stated re- sponse can be implemented. They also examine whether the policy’s scope is adequate. Separate categories pre- serve protections for affected people and the develop- ment side of AI policy. Every sub-criterion requires a cited provision and an ordered judgment. Category and criterion results show how the final score was reached. The coverage report identifies missing evidence, while the instrument list and cutoff make later rescoring pos- sible. The result is a method for evaluating documented de- sign and implementation readiness. A strong score does not establish that an institution has enforced the policy or that a regulated organization complies with it. It also does not prove an improvement in safety or individual rights. Innovation and public welfare require outcome evidence of their own. The next empirical step is to ap- ply one frozen version of the rubric to complete port- folios using independent evaluators. Publishing the ev- idence and initial judgments will allow reliability and sensitivity to be tested. The instrument should then be revised only where application reveals a genuine ambi- guity or omission. A Compute Infrastructure Review This thirteen-jurisdiction inventory informed the purposive selection in Section 2.1. It provides context only and does not enter the policy score. No.CountryScientific Supercomputing Fa- cilities Private Company ClustersCloud Compute Regions 1ThailandThe TARA system reports a theo- retical peak near 500 teraFLOPS from 4,136 Intel Xeon CPU cores and 28 NVIDIA Tesla V100 GPUs (Sakdhnagool et al. 2021). Siam.AI and other firms have an- nounced private AI infrastructure investments(Leesa-nguansuk 2025). Alibaba Cloud and other large providers offer cloud services in Thailand (Zulhusni 2025). 2MalaysiaYTL’s AI Cloud in Johor uses NVIDIA systems for commercial and research workloads (YTL Group 2024). YTL and cloud providers have announced private data-center andacceleratorcapacityin Malaysia (YTL Group 2024). AWS and Microsoft have an- nounced or opened Malaysian cloud regions (Ines Lin 2024). 3VietnamViettel and research institutions have developed computing facil- ities for scientific and AI work- loads (Viettel 2022). Vingroup supports AI research and data infrastructure through VinAI and VinBigdata (Merritt 2021; VinAI 2024). Foreign and domestic providers offer cloud services, while the cited AWS presence was an edge location rather than a full region (AWS 2022). 4PhilippinesThe National Center for AI Re- search was established to support applied research and public- sectorprojects(dtiwebteam 2024). Converge ICT and Supermi- cro announced data-center in- frastructure for AI workloads (Loyola 2024). The reviewed cloud source de- scribes an expanding domestic data-center market (Karen 2022). 5IndonesiaIndosat and NVIDIA announced AI cloud infrastructure using high-density accelerator systems (Wire 2024; NVIDIA 2022). GoTo and other technology firms operate private infrastructure for consumer and business services (Baskoro 2024). AWS, Google Cloud, and Mi- crosoft operate or have an- nounced cloud regions in Indone- sia (Mehra 2024). 6SingaporeThegovernmentcommitted S$270 million to expand the National Supercomputing Centre and related training (Keat 2024). Commercial providers operate private cloud and accelerator ca- pacity in Singapore (G 2025). AWS, Microsoft Azure, and Google Cloud operate Singapore cloud regions (G 2025). 7JapanABCI 3.0 became publicly avail- able in January 2025 with 6,128 NVIDIA H200 GPUs and a stated peak of 6.22 EFLOPS at FP16 (NVIDIA 2025a; Takano et al. 2024). SoftBank, Fujitsu, NEC, and other firms operate or have an- nounced private compute clusters (Narioka 2024). Domestic and global providers operate Japanese cloud regions. The comparison does not equate commercial regions with scien- tific supercomputers. 8South KoreaSamsung and SK Hynix sup- ply memory used in AI systems, while the national infrastructure plan includes public and private compute capacity (Park 2024; NVIDIA 2025b). Naver, Kakao, Samsung, and SK Hynix operate research or compute facilities (Chang-won 2021). AWS, Google Cloud, Microsoft Azure, KT Cloud, and NHN Cloud offer services in South Ko- rea (Butler 2025). 9IndiaThe PARAM series and National Supercomputing Mission pro- vide public high-performance computing capacity (Department ofScienceandTechnology 2022). Large technology and telecom- munications firms operate private infrastructure for AI and data ser- vices (Kumar 2024). AWS, Google Cloud, Microsoft Azure, and Oracle operate In- dian cloud regions (APAC Media 2023). 10UKARCHER2,CSD3,and Isambard-AIprovidescien- tific and AI computing capacity (Research and Innovation 2021; Shainer 2021; University of Bristol 2025). London,Cambridge,Oxford, Manchester,andEdinburgh host commercial AI firms and research clusters (Korolov 2025). AWS, Microsoft Azure, Google Cloud, and specialist providers operate UK data centers and cloud services (Jackson 2025). Continued on next page Table 4 – continued from previous page No.CountryScientific Supercomputing Fa- cilities Private Company ClustersCloud Compute Regions 11USAOn the June 2026 TOP500 list, El Capitan, Frontier, and Aurora ranked second through fourth be- hind China’s LineShine system (TOP500.org 2026b). The Na- tional AI Research Resource pi- lot provides shared resources for academic and public-interest re- search (Miller and Gelles 2024). Commercial capacity includes NVIDIAGPUclustersand custom accelerators such as Google’s Ironwood TPU, AWS Trainium, and Cerebras wafer- scale systems (NVIDIA 2026; Google Cloud 2026; Amazon Web Services 2026; Cerebras Systems 2026). AWS, Google Cloud, and Mi- crosoft Azure operate multiple US regions with GPU and proprietary accelerator services (Lehdonvirta2024;Amazon Web Services 2026; Google Cloud 2026). 12European Union EuroHPCcoordinatesshared public supercomputing systems, includingLUMIinFinland and facilities in Germany and France (European Union 2021a; HPCwire 2024; Krause and Th ̈ ornig 2016; CNRS 2025). Europeantelecommunications and technology firms operate private data centers, and global firms maintain facilities in sev- eral member states (Atos 2020; Williams 2023; Meta 2023). AWS, Google Cloud, and Mi- crosoft Azure hold a large share of the European cloud market, with regions across several mem- ber states (Goovaerts 2022). 13ChinaLineShine at the Shenzhen Na- tional Supercomputing Center ranked first on the June 2026 TOP500 list at 2.198 exaFLOPS (TOP500.org 2026a). Huawei’s Atlas 900 A3 Super- PoD connects up to 384 Ascend chips (Huawei 2025). AlibabaCloudandTencent Cloud document regions across China. The East Data West Computing program coordinates national computing hubs and data-centerclusters(Alibaba Cloud 2026; Tencent Cloud 2026; State Council of the People’s Republic of China 2024a). Table 4: Infrastructure for AI Compute by Country The table below records selected economic, political, and external conditions from sources that use different measures. The entries provide context and are not a comparable stability ranking. No.CountryGDPandEconomic Growth Political and External Context RegulatoryFrame- works Resource Availability 1ThailandThe Eastern Economic Corridorandnational plans identify advanced technology as an eco- nomic priority (under the Ministry of Digital Econ- omy and MDES). The national AI strategy and Eastern Economic Corridorprovidethe selected policy context (under the Ministry of DigitalEconomyand MDES). Industry 4.0 programs operatealongside data-protectionand cybersecurityrules (under the Ministry of Digital Economy and MDES). NSTDA supports techni- cal training and research (NSTDA 2019). 2MalaysiaThe economy and na- tional development plans supportinvestmentin digitalinfrastructure (IMF 2025a). Thecitedinvestment analysisandthe NVIDIA-YTLproject show continued foreign anddomesticinfras- tructurecommitments (Hattangudi, Tharar, and Kukreja2024;MIDA 2024b). Data-protection, con- sumer,transaction, andcybersecurity rules form the relevant regulatory base (ITA 2024). Solar projects and the YTL Green Data Cen- ter add power and data- center capacity (MIDA 2024a; YTL 2022). 3VietnamRapid economic growth andpublicdigital- infrastructure programs supporttechnology investment(FocusEc- onomics2024;Hanh 2024). Nationaldigital- transformationpolicy providestheselected policycontext(US DepartmentofState 2024). TheNationalDig- italTransformation Roadmapoperates alongside developing data and AI rules (Hoa 2025; Samuel 2021). Reviewedsources document foreign part- nerships and domestic researchinstitutions, but do not provide a comparable estimate of high-end compute access (NVIDIA 2024; Tazrout 2021). Continued on next page Table 5 – continued from previous page No.CountryGDPandEconomic Growth Political and External Context RegulatoryFrame- works Resource Availability 4PhilippinesThe cited forecast re- portedpost-pandemic growth near 5–6% (IMF 2025b). No comparable quanti- tative stability measure was retained for this re- view. TheNationalAI Roadmap 2.0 supports research,adoption, andresponsible-use guidance (DTI 2024). Renewable-energy devel- opment can support data- center expansion, though the cited sources do not quantify AI-specific sup- ply (Tachev 2024; IMF 2020). 5IndonesiaThe cited forecast placed thedigitaleconomy above $130 billion by 2025 (ITA 2025). No comparable quanti- tative measure was re- tained for this field. ThePersonalData Protection Law and national ethics guid- ance provide the main horizontal rules rele- vant to AI (Deradjat et al. 2025; Digital Policy Alert 2023). The cited source de- scribes domestic energy andinfrastructurere- sources but does not isolateAIuse(East Ventures 2025). 6SingaporeThe economy grew 4.8% in 2025. Budget mea- sures committed S$1 bil- lion over five years to AI and S$150 million to the Enterprise Compute Ini- tiative (Ministry of Trade and Industry, Singapore 2026; The Edge Singa- pore 2024; Mothership 2025). No comparable quanti- tative measure was re- tained for this field. Data-protectionand cybersecurityrules operatealongside nonbinding AI frame- works. A dedicated agentic AI framework followed in January 2026(Baig,Khan, andGardezi2024; Infocomm Media De- velopment Authority, Singapore 2026). Singtel and the National Supercomputing Centre provide commercial and public compute capac- ity (Gomes 2024; Chia 2024). 7JapanA large economy and sustained technology in- vestment support public and private AI programs (Wolf 2025). G7 participation and in- ternational digital-policy partnerships shape the external setting (Min- istry of Economy and Industry 2022). Human-centered prin- ciplesandsectoral guidancenowop- eratealongsidethe 2025 AI Promotion Act (Cabinet Office, Government of Japan 2025; Government of Japan 2025). ABCI and the private clusterslistedabove arethedocumented resources used in this comparison. 8South KoreaNominal GDP was about $1.86–1.87trillionin 2025. The government also announced large AI financing and compute programs (International Monetary Fund 2026; Min-gwan and Su-hyeon 2025; NVIDIA 2025b). Trade relationships and semiconductorsupply- chain partnerships shape theexternalsetting (Administration 2023). The AI Basic Act took effect on January 22, 2026. It distin- guisheshigh-impact AIbysectorand consequencefrom high-performance AI defined partly by a 10 26 operation thresh- old (Republic of Korea 2025; Jon 2026). Samsung and SK Hynix providesemiconductor manufacturing capacity relevant to AI systems (Administration 2023). Continued on next page Table 5 – continued from previous page No.CountryGDPandEconomic Growth Political and External Context RegulatoryFrame- works Resource Availability 9IndiaNominal GDP was es- timated at about $4.15 trillion in 2025, with 6.6% real growth pro- jected for fiscal year 2025–26(International Monetary Fund 2025). No comparable quanti- tative measure was re- tained for this field. The national strategy and IndiaAI Mission operate alongside the 2023data-protection statute,2025rules, and2025volun- tary AI Governance Guidelines(Press InformationBureau, Government of India 2025;Ministryof Electronics and Infor- mationTechnology, Government of India 2025). Nationalskillingpro- grams, research institu- tions, and commercial cloud services support AI development (NeGD 2020). 10UKNominal GDP reached about £3.03 trillion in 2025. The government adopted the AI Opportu- nities Action Plan, and firms announced invest- ment under the 2025 technologyagreement (HouseofCommons Library 2026; Clifford andDepartmentfor Science,Innovation and Technology 2025; CNBC 2025). G7andotherinter- nationalpartnerships supportresearchand policycoordination (for Data Ethics and Innovation 2021). The Data Protection Act and sectoral law operate alongside non- statutory AI principles. The AI Security In- stitute and AI Growth Labaddevaluation and sandbox functions (Parliament2018; Department for Sci- ence, Innovation and Technology2025; Rough and House of CommonsLibrary 2026). Public research funding, Innovate UK, and pri- vate investment support compute and talent (ITA 2023; UKRI 2022). 11USANominal GDP was about $30.5–30.8trillionin 2025. Large technology firms directed substantial capital spending toward data centers and spe- cialized hardware (U.S. BureauofEconomic Analysis 2026). Export controls on ad- vanced chips changed several times in 2025, in- cluding rescission of the AI Diffusion Rule before it took effect (Bureau of Industry and Security, U.S.Departmentof Commerce 2025; RFA 2025). ExecutiveOrder 14179 and the July 2025 AI Action Plan replacedtheprior executive-order frame- work. The CHIPS Act continues to support domestic semiconduc- tor production (The White House 2025b; The White House / Office of Science and TechnologyPolicy 2025; Kurilla 2025). The public and private fa- cilities listed above are the documented resource base used in this compar- ison. 12European Union GDP wasC16.22 tril- lion in 2024. InvestAI targetsC200billion in public and private AIinvestment,while 2025 private investment remained well below the US level (Eurostat 2026; EuropeanCommission 2025b; Stanford Institute forHuman-Centered ArtificialIntelligence 2026). As a supranational com- parator, the EU entry describescollective digital policy rather than a single national political environment(Gaia-X EuropeanAssociation for Data and Cloud 2025; EuropeanCommission 2026). The AI Act applies in phases. Regulation 2026/1744revised parts of the imple- mentationschedule (EuropeanCommis- sion 2024b; European Union 2026). EuroHPC, the Connect- ing Europe Facility, and member-stateenergy systems support infras- tructure across the Union (Commission 2021). Continued on next page Table 5 – continued from previous page No.CountryGDPandEconomic Growth Political and External Context RegulatoryFrame- works Resource Availability 13ChinaFinal revised GDP for 2024 was 134.81 trillion yuan, with reported real growth of 5.0% (National Bureau of Statistics of China 2025; Zhou and Interesse 2025). The 2024 Global Peace Index ranked China 88th of 163 jurisdictions (IEP 2024). The 2017 development plan, 2025 AI Plus initiative,synthetic- content labeling rules, and amended Cyberse- curity Law form part of the national port- folio (Webster et al. 2017; State Council of the People’s Re- public of China 2025; Cyberspace Adminis- tration of China and others 2025; National People’sCongress StandingCommittee 2025). The East Data West Computing program di- rects national computing hubsanddata-center clusters toward eastern andwesternregions (State Council of the People’sRepublicof China 2024a). Table 5: Economic and Political Environment by Country The entries below summarize signals reported by the cited sources. They do not establish causal effects or provide a common measure of demand or talent across jurisdictions. No.CountryLocal Demand for AI solutionsResearch InstitutionsTalent Pool 1ThailandThe cited reporting describes in- vestment in domestic AI services and interest among Thai busi- nesses (Leesa-nguansuk 2025). NSTDA and universities sup- port AI research and collabo- ration with industry (Dharmaraj 2025b). Government and university pro- grams provide training in AI and data science (Dharmaraj 2025b). 2MalaysiaReviewed sources describe AI adoption in finance, healthcare, and manufacturing as part of wider digitalization efforts (ITA 2024). The Malaysia Genome Institute and MRANTI support research and commercialization in ad- vanced technologies (Malaysi- akini 2024). Education programs seek to ex- pand the workforce, while re- ported shortages remain in AI, data analytics, and cybersecurity skills (Vietnamplus 2025). 3VietnamPhoBERT documents work on Vietnamese-languagemodels, while the reviewed sources also describe applications in health- care, transportation, and finance (Dat Quoc Nguyen 2020). VinAI and university research groups conduct AI research and collaborate with overseas institu- tions (Nhandan 2025). Universities and companies fund AI training, while the reviewed sourcereportsshortagesin advanced expertise (Dharmaraj 2025a). 4PhilippinesThe national AI roadmap and the National Center for AI Research identify healthcare, agriculture, and manufacturing as areas for AI development (Dti 2024). The roadmap links the center with universities and industry partners for research and knowl- edge transfer (Dti 2024). The roadmap includes reskilling and workforce development in- tended to address reported talent shortages (Dti 2024). 5IndonesiaThe reviewed report identifies fintech, electronic commerce, and healthcare as areas of AI adoption and places AI within the “Making Indonesia 4.0” agenda (Ministry of Industry 2018). Bandung Institute of Technology, the University of Indonesia, and other institutions conduct AI and data-science research (Ministry of Industry 2018). The report identifies gaps in workforce readiness and de- scribes public programs in digital skills and STEM education (Min- istry of Industry 2018). 6SingaporeThe National AI Strategy identi- fies finance, healthcare, logistics, and public services as areas for wider AI use (Smart Nation Sin- gapore 2019). The National University of Sin- gapore, Nanyang Technological University, and public research organizations support research andindustrycollaboration (Smart Nation Singapore 2019). The strategy sets a target of tripling the number of AI prac- titioners to 15,000 by 2028 and describes scholarships and train- ing programs (Smart Nation Sin- gapore 2019). Continued on next page Table 6 – continued from previous page No.CountryLocal Demand for AI solutionsResearch InstitutionsTalent Pool 7JapanThe reviewed market report iden- tifies manufacturing, healthcare, and robotics as major areas of AI adoption (Insights 2023). The University of Tokyo, AIST, and other institutions use the AI Bridging Cloud Infrastructure for research (Harris 2024). Government and university pro- grams support AI education and research training (Harris 2024). 8South KoreaThe reviewed report describes AI use in healthcare, finance, man- ufacturing, urban planning, and public services (McFaul et al. 2023). KAIST, Seoul National Univer- sity, and other institutions con- tribute to AI research and patent- ing (McFaul et al. 2023). National programs fund AI ed- ucation and workforce training in response to reported shortages (McFaul et al. 2023). 9IndiaReviewed sources describe AI use and demand in healthcare, agriculture, education, and urban services, alongside public digi- talization programs and startup investment (Singh et al. 2025; Singhania 2024). IITs, IIITs, NITs, public agen- cies, and research hubs conduct AI research and support collab- oration with industry (Peterson 2024; Arnab Kumar and Mahin- dru 2024; Ministry of Education India 2022). National and state programs pro- vide AI training, including “Re- sponsible AI for Youth,” while industry sources continue to re- port unmet demand for advanced skills (Stanly 2024; drishtiias 2020). 10UKThe cited sources document AI activity in healthcare, financial services, and urban applications (Benn 2025). Oxford, Cambridge, Imperial CollegeLondon,University College London, and the Alan TuringInstituteconductAI research and collaborate with government and industry (Benn 2025; Sharma 2024). UKRI doctoral funding, com- panypartnerships,andthe Global Talent Visa form part of the country’s AI training and recruitment system (Benn 2025). 11USAThe 2025 AI Index reports ex- tensive private investment and commercial model development, while public procurement in- cludes defense and health appli- cations (Maslej et al. 2025b; AP 2023). Universities and federally funded AI institutes support research and collaboration with industry (Maslej et al. 2025b). The United States attracts re- searchers and engineers from abroad, while immigration policy affects recruitment and retention (Oschinski et al. 2025). 12European Union EU institutions and member- state programs support AI adop- tion across manufacturing, fi- nance, healthcare, and public ser- vices (Eurostat 2025). Universities,nationallabora- tories, and the ELLIS network supportAIresearchacross member states (Tambiama Mad- iega with Rafał Ilnicki; Graphics: Eulalia Claros 2024). Universities and digital-skills programs train AI workers, but available sources use different occupationaldefinitionsand should not be combined into one workforce count (Allgeyer 2023). 13ChinaBaidu, Alibaba, Tencent, govern- ment agencies, and state-owned enterprises procure or provide AI services in domestic markets (CKGSB 2025). Supercomputing centers, univer- sities, research institutes, and company laboratories contribute to AI research (Yang 2024). The 2025 AI Index reports that China produced 23.2% of AI publications in 2023 (Stanford Institute for Human-Centered ArtificialIntelligence2025). MacroPolo reports that 11% ofChina-educatedeliteAI researchers in its 2024 and 2025 conference sample worked in China, compared with 16% in its 2019 sample (MacroPolo 2025). Table 6: Demand and Talent in AI by Country B Cross-National Policy Overview The tables report the status of the reviewed instruments through August 15, 2026. They describe what the cited sources state and do not infer causal effects. PolicyInstru- ment Covered ActorsDocumented ProvisionsImplications for AI Sys- tems ImplementationStatus and Limitations European Union ArtificialIn- telligenceAct (2024)(Eu- ropeanUnion 2021b;Euro- pean Parliamen- taryResearch Service 2022) AI Developers, Businesses, PublicSector, End Users 1. Prohibits specified AI practices 2. Imposesdutieson providers and deployers of high-risk systems 3. Adds transparency du- ties for specified sys- tems 4. Regulatesgeneral- purpose AI models 5. Providesconformity assessment,documen- tation, oversight, and penalties 1. High-risk systems must meet applicable risk- management, data, doc- umentation,oversight, accuracy,robustness, and cybersecurity duties 2. Specified practices are prohibited 3. General-purposeAI providers have separate documentation and risk duties 1. Entered into force in August 2024 2. Prohibited-practice and general-purpose AI pro- visions began applying in February and August 2025 (European Com- mission 2024b) 3. Regulation2026/1744 revised several appli- cation dates (European Union 2026) GeneralData Protection Reg- ulation (GDPR, 2018)(Euro- peanUnion 2016) Organizations, AI Developers, Individuals 1. Requires a lawful basis for processing personal data 2. Data minimization 3. Rightsconcerning solely automated deci- sions 4. Access to meaningful information about the logic involved where ap- plicable 5. Conditions for transfers outside the European Economic Area 1. Applies when model de- velopment or use pro- cesses personal data 2. Truly anonymized data falls outside the Regula- tion 3. Article 22 limits spec- ified decisions based solelyonautomated processing 1. Binding and directly ap- plicable since 2018 2. Enforced by national su- pervisory authorities 3. Does not provide a gen- eral AI approval regime DigitalSer- vicesAct (DSA,2022) (European Commission 2022) OnlinePlat- forms,Con- sumers, Regula- tors 1. Recommender-system transparency 2. Advertisingtrans- parency 3. Systemic-riskassess- ment and mitigation for very large services 4. Notice, complaint, and redress procedures for content moderation 1. Applies to platform sys- tems used for recom- mendation, advertising, and content moderation 2. Requires risk mitigation by very large online platforms and search en- gines 3. Does not impose a gen- eral ban on models that amplify harmful content 1. Binding platform regu- lation 2. Obligations depend on service type and size 3. Enforced by the Com- mission and national Digital Services Coordi- nators Table 7: European Union AI Policy Overview PolicyInstru- ment Implementing Body Purpose and ProvisionsImplications for AI Sys- tems ImplementationStatus and Limitations Executive Order 14110(2023), revoked January 20, 2025 (White House2023; TheWhite House 2025a) Executive Office of the President 1. Directed federal action on safety, security, pri- vacy, civil rights, re- search, and workforce issues 2. Used Defense Produc- tion Act authority for specified reporting 3. Directedagenciesto develop standards and guidance 1. Required specified de- velopers to report train- ing and safety informa- tion 2. Directedevaluations andstandardsfor severalhigh-impact uses 1. Revoked on January 20, 2025 2. Nolongerforms theoperativefed- eralexecutive-order framework Executive Order 14179(Jan. 2025) & Amer- ica’s AI Action Plan (July 2025) (TheWhite House2025b; TheWhite House / Office of Science and Technology Policy 2025) Executive Office of the President andfederal agencies 1. Reorients federal policy toward US AI leadership 2. Directs revision of the NIST AI RMF 3. Renames the AI Safety Institute as the Center for AI Standards and In- novation (CAISI) 4. Organized around in- novation, infrastructure, and international diplo- macy 1. Removes or revises sev- eral prior federal re- quirements 2. Directs work on infras- tructure, standards, pro- curement, exports, and security evaluations 1. Operative federal policy at the cutoff 2. The AI Diffusion Rule was rescinded in May 2025 before taking ef- fect 3. State AI laws remain outside this federal port- folio AIBillof Rights Blueprint (2022)(White House 2022) WhiteHouse OSTP 1. Nonbindingprinciples on safe systems, dis- crimination,privacy, notice, explanation, and human alternatives 2. Includes technical com- panion guidance 1. Providesdesignand governanceguidance for automated systems 2. Creates no approval re- quirement or penalty 1. Published as a White House blueprint 2. Nonbinding and not a source of legal authority NationalAI InitiativeAct (2020)(U.S. Congress 2020) US Congress 1. Establishesacoor- dinatedfederalAI research and develop- ment program 2. Supports research insti- tutes, education, stan- dards, and interagency coordination 1. Authorizesprograms and coordination rather thanmodel-specific approval rules 1. Binding statutory basis for federal programs 2. Doesnotcreatea comprehensive private- sector AI regime AI Risk Man- agement Frame- work(NIST AIRMF) (2023)(Na- tionalInstitute ofStandards and Technology (NIST) 2023) NIST 1. Voluntaryframework for managing AI risks across the lifecycle 2. Organized around Gov- ern, Map, Measure, and Manage functions 1. Providesacommon risk-managementvo- cabulary and process 2. Creates no independent approval requirement 1. Voluntaryunless anotherinstrument incorporates it 2. Does not itself autho- rize penalties or suspend systems Continued on next page Table 8 – continued from previous page PolicyInstru- ment Implementing Body Purpose and ProvisionsImplications for AI Sys- tems ImplementationStatus and Limitations California Con- sumerPrivacy Act and Cali- forniaPrivacy RightsAct (California State Government 2023) California State Government 1. Grants access, deletion, correction, and opt-out rights for covered per- sonal information 2. Adds limits and rights for sensitive personal in- formation 1. Applies to covered busi- nesses that process per- sonal information in AI development or use 2. Does not create a gen- eralAItraining-data consent rule 1. Binding state law with threshold and exemption rules 2. Enforced by the Cal- ifornia Privacy Protec- tion Agency and Attor- ney General 3. Not part of the fed- eral portfolio score un- less the declared scope includes subnational law Table 8: United States AI Policies Overview PolicyInstru- ment Implementing Body PurposeDocumented ProvisionsImplementationStatus and Limitations Artificial Intelli- gence in Health- care Guidelines (AIHGle) Ministryof Health (MOH), HealthSci- ences Authority (HSA),Inte- gratedHealth Information Systems (IHiS) 1. Guide safe development and use of AI in health- care 1. Recommendsvali- dation,monitoring, fairness, responsibility, explainability,and patient-centered design (MinistryofHealth Singapore,Health SciencesAuthority, and Integrated Health InformationSystems 2023) 2. Addresses data gover- nance and cybersecurity 1. Guidance rather than a standalone approval statute 2. Separate medical-device and professional rules can create binding du- ties PersonalData ProtectionAct (PDPA)and Sector-Specific Guidelines PersonalData Protection Commission (PDPC) 1. Protect personal data processed by organiza- tions 2. Provide nonbinding AI governance guidance 1. Requirescompliance with consent, purpose, notification,access, correction,protection, retention, and transfer duties where applicable 2. The Model AI Gover- nance Framework pro- vides separate organi- zational guidance (Per- sonal Data Protection Commission Singapore 2020) 1. The PDPA is binding and technology neutral 2. The Model AI Gov- ernance Framework is nonbinding MASFEAT Principles Monetary Authorityof Singapore (MAS) 1. Promotefairness, ethics,accountability, andtransparencyin AI use within financial sector 1. Setsfairness,ethics, accountability,and transparency principles (Monetary Authority of Singapore 2021) 2. Supportsassessment methods for financial institutions 1. Sectoralsupervisory guidance rather than a general AI statute Continued on next page Table 9 – continued from previous page PolicyInstru- ment Implementing Body PurposeDocumented ProvisionsImplementationStatus and Limitations Healthcare Cy- bersecurity Es- sentials Guide- lines Ministryof Health (MOH) 1. Protect healthcare AI systems and IT infras- tructure against cyber threats 1. Recommendsaccess control,protection, detection,response, recovery,andstaff practices (Ministry of Health Singapore 2022) 1. Generalhealthcare cybersecurityguid- ance rather than an AI-specific rule Human Biomed- icalResearch Act (HBRA) Ministryof Health (MOH) 1. Ensure ethical compli- ance and data privacy in biomedical AI research 1. Requires review, con- sent, and protection for covered human biomed- ical research (Ministry ofHealthSingapore 2007) 1. Bindingforresearch withinitsstatutory scope 2. Not a general approval regime for AI models or clinical products Table 9: Overview of Singapore AI Policies Policy Instru- ment Covered ActorsDocumentedProvi- sions Implications for AI Systems ImplementationStatus and Limitations UKIPOAI and IP Con- sultationand Government Response, 2021 and 2022 (UK Intellectual Property Office 2022) 1. Creators,rights holders, researchers, AI developers, and inventors 1. Considerscopy- right protection for computer-generated works,textand data mining, and patentprotection forAI-devised inventions 2. The 2022 response retainedexisting protectionfor computer-generated worksanddid not change patent inventorship rules 3. The response pro- posed a broader text and data mining ex- ception, which was not implemented 1. Theconsultation and response ad- dress how existing intellectual property law applies to AI developmentand output 2. They do not amend the underlying law 1. Consultation opened in October 2021 and the governmentresponse was published in June 2022 2. This is intellectual prop- erty policy rather than an AI safety or approval regime UK AI White Paper(UK Department forScience, Innovation and Technology 2023b) 1. Existing regulators and the organiza- tions within their sectoral remits 2. AI developers and deployers whose ac- tivities fall within those remits 1. Uses existing law and existing regula- tors rather than a single horizontal AI regulator 2. Setsfivecross- sectorprinciples coveringsafety, transparency,fair- ness, accountability, and contestability 3. Proposes central co- ordination and sup- port for regulators 1. Regulators apply the principles to AI uses within their existing powers and sectoral context 2. Theframework does not create a generallicensing orconformity- assessment regime 1. Nonstatutorypolicy frameworkwhose applicationinitially dependedoneach regulator’s remit and discretion 2. TheOctober2025 regulatory blueprint re- tained the regulator-led approach (Rough and HouseofCommons Library 2026) Continued on next page Table 10 – continued from previous page Policy Instru- ment Covered ActorsDocumentedProvi- sions Implications for AI Systems ImplementationStatus and Limitations NationalAI Strategy(UK Government 2021) 1. Government,re- search institutions, businesses, and the workforce 1. Sets ten-year priori- ties for investment, research,skills, adoption,and governance 2. Organizesthe strategyaround ecosystemneeds, diffusionacross the economy, and national and inter- national governance 1. Providesstrategic direction for public programs and later policy instruments 2. Does not create con- duct duties for indi- vidual AI systems 1. Published in September 2021 as a strategy rather than legislation 2. The 2025 AI Opportu- nities Action Plan sup- plements its infrastruc- ture, adoption, and skills agenda AI Opportuni- ties Action Plan (Cliffordand Department forScience, Innovation and Technology 2025) 1. UK government and delivery bodies 2. Regulators and pub- lic sector organiza- tions 3. AI developers, re- searchers, and in- vestors 1. RecommendsAI Growth Zones and expanded sovereign compute 2. Proposes better ac- cess to public data and stronger adop- tion capacity 3. Addressesskills, research,pro- curement, and AI security institutions 1. Expands the infras- tructure and institu- tions available for AI development and use 2. Does not itself set bindingconduct rules for particular model classes 1. Published and accepted by the government in January 2025 2. Implementationde- pends on later budgets, programs, and institu- tional decisions 3. The plan is a strategy document rather than legislation UKAl- gorithmic Transparency Standard (UK Government Digital Service 2025) 1. All government de- partments 2. Arm’s-length bodies that deliver public or frontline services or interact directly with the public 3. Other public-sector bodies can use the standard voluntarily 1. Providesastan- dard record for an algorithmictool’s purpose,owner- ship,data,risks, humanoversight, and review 2. Requirespubli- cation of records forcoveredor- ganizationsand tools 1. Mandatoryscope coverstoolsthat significantlyin- fluence a decision with public effect or interact directly with the public 1. Mandatory across cen- tral government since 2024 2. Thescopeandex- emptionspolicywas published in December 2024 and the guidance was updated in May 2025 Table 10: United Kingdom AI Policy Portfolio Policy Instru- ment Covered ActorsDocumentedProvi- sions Implications for AI Systems ImplementationStatus and Limitations National Strategyfor Artificial Intel- ligence,2018 (NITIAayog 2018) 1. Government depart- ments 2. Healthcare,agri- culture,education, urban services, and mobility sectors 3. Startups and small and medium enter- prises 4. Researchinstitu- tions 1. Identifies five pri- oritysectorsand an inclusive-growth objective 2. Proposesresearch centers, shared data platforms, skilling, andpublic-private collaboration 3. Discussesprivacy, security,trans- parency,account- ability,andbias asresponsible-AI concerns 1. Sets priorities for public programs and sector adoption 2. Doesnotcreate approval, audit, or penalty rules for AI systems 1. Nonbindingnational strategyratherthan legislation 2. Later programs and the 2025 AI Governance Guidelines supplement itsdevelopmentand governance proposals DigitalPer- sonalData Protection Act 2023 and Rules2025 (Government of India 2023b; Press Informa- tionBureau, Government of India 2025) 1. Data fiduciaries and processors 2. People whose digi- tal personal data are processed 3. The Data Protection Board and central government 1. Requires a lawful purpose based on consent or a listed legitimate use 2. Provides rights to information, correc- tion, erasure, and grievance redress 3. Sets additional du- ties for significant data fiduciaries 4. Permitsthegov- ernment to restrict transferstonoti- fiedcountriesor territories 1. Applies when AI training or deploy- mentprocesses digital personal data within its scope 2. Does not create a separate approval or audit regime for AI systems 1. The 2025 Rules began a staged implementation of the Act 2. Several substantive du- ties commence on later dates under that sched- ule 3. The regime is bind- ing data protection law rather than AI-specific regulation Revised MeitY Advisory on AI Deployment, March15, 2024 (AZB and Partners 2024) 1. Intermediaries and online platforms 2. Providers that make AI tools available through those ser- vices 1. Links AI use to ex- isting intermediary due diligence obli- gations 2. Directsplatforms to address unlawful content, discrimina- tion, and electoral integrity 3. Calls for warnings onoutputsfrom under-testedor unreliable systems 4. Addresses identifi- cation of synthetic content used as mis- information or deep- fakes 1. Doesnotrequire priorgovernment permissionto release an under- tested model 2. Places the stated du- ties on intermedi- aries and platforms rather than all AI de- velopers 1. Superseded the March 1 advisory that had re- quired prior permission 2. Its scope and legal force remain less clear than those of legislation or notified rules Continued on next page Table 11 – continued from previous page Policy Instru- ment Covered ActorsDocumentedProvi- sions Implications for AI Systems ImplementationStatus and Limitations India AI Gov- ernance Guide- lines,2025 (Ministryof Electronics and Information Technology, Government of India 2025) 1. Central and state government bodies 2. Sector regulators 3. AI developers, de- ployers, and users 1. Setsprinciples fortrustworthy, people-centered, andinnovation- supportinggover- nance 2. Recommendsa coordinatinggov- ernance group and a technical policy committee 3. Recommendsrisk assessment,inci- dent reporting, and sharedtechnical tools 1. Uses existing law and sector oversight as the base for AI governance 2. Supportspropor- tionateobligations andregulatory sandboxes 1. Issued as national guid- ance in November 2025 2. Recommendationsre- quire later institutional andsector-specific implementation 3. The document does not itself create a compre- hensive AI statute Table 11: India AI Policy Portfolio Policy Instru- ment Covered ActorsDocumentedProvi- sions Implications for AI Systems ImplementationStatus and Limitations ActonPro- motionof Researchand Development and Utilization ofAI-Related Technologies, 2025 (Cabinet Office, Govern- ment of Japan 2025) 1. National and local government 2. Researchinstitu- tions and businesses 3. AIdevelopers, providers, and users 1. Establishes an AI Strategy Headquar- ters and a national basic plan 2. Assigns responsibil- ities for research, adoption, safety, and international coop- eration 3. Authorizes informa- tion requests, guid- ance, and public dis- closure when mis- use harms rights or interests 1. Relies mainly on planning,guid- ance,voluntary cooperation,and disclosure 2. Doesnotcreate agenerallicens- ing regime for AI models 1. Promulgated on June 4, 2025 2. Fully in force from September 1, 2025 3. Doesnotprescribe private-sector fines or criminal penalties for noncompliance Society5.0 Vision (Cabinet Office, Govern- ment of Japan 2019) 1. Government,re- searchers,and industry 2. Publicservice andinfrastructure sectors 1. Defines a human- centered society that integrates physical and digital systems 2. Connectstechno- logical development with economic and social goals 1. Providesstrategic directionforAI and related digital technologies 1. A high-level national vi- sion rather than a bind- ing conduct standard SocialPrinci- ples of Human- CentricAI, 2019 (Japanese Ministryof Economy, Tradeand Industry 2020) 1. Government, busi- ness,researchers, and the public 1. Addresseshuman dignity,educa- tion,privacy, security, fair com- petition,fairness, accountability, transparency,and innovation 1. Supplies principles for later strategies and business guid- ance 1. Nonbindingprinciples with no direct penalty mechanism Continued on next page Table 12 – continued from previous page Policy Instru- ment Covered ActorsDocumentedProvi- sions Implications for AI Systems ImplementationStatus and Limitations AI Guidelines forBusiness, Version1.0 (Ministryof Economy, Tradeand Industryand Ministryof Internal Affairs and Communi- cations, Japan 2024) 1. AIdevelopers, providers,and business users 1. Consolidates earlier guidanceforAI governance and risk management 2. Addressessafety, fairness, privacy, se- curity, transparency, and accountability 1. Provideslifecycle guidance that organ- izations can adapt to their roles and risks 1. Publishedjointlyby METI and MIC in April 2024 2. Guidance rather than binding legislation Table 12: Japan AI Regulations Overview PolicyInstru- ment Covered ActorsDocumented ProvisionsImplications for AI Sys- tems ImplementationStatus and Limitations Framework Act on the Develop- ment of Artifi- cial Intelligence and the Creation of a Foundation for Trust (Re- public of Korea 2025; Jon 2026) Government bodies, AI busi- ness operators, andpeople affected by AI services 1. Establishesnational planning and coordina- tion institutions 2. Sets transparency duties for generative and high- impact AI 3. Requiressafety measuresforhigh- performance AI 4. Setssafeguardsfor high-impact AI and en- courages fundamental- rights impact assess- ment 1. High-impact AI is iden- tified by sector, use, and risk to life, safety, or fundamental rights 2. High-performance AI is a separate class tied in part to training compute of at least 10 26 opera- tions 3. Covered systems remain deployable when the ap- plicable duties are met 1. Enacted on January 21, 2025 and effective from January 22, 2026 2. Subordinaterules specifyoperational details and the high- performancecompute threshold 3. The Act combines bind- ing duties with promo- tion and support mea- sures Personal Infor- mation Protec- tion Act, revised 2023 (Personal Information Protection Commission, Republicof Korea 2023) Personalin- formation controllers and data subjects 1. Sets lawful-basis, pur- pose,security,and data-subject-rights requirements 2. Article 37-2 provides rights concerning qual- ifying decisions made through a completely automated system 3. Provides a right to re- quest an explanation of an automated decision 1. Applies when AI de- velopment or use pro- cesses personal infor- mation within the Act’s scope 2. Requires specified re- sponses when a person objects to or requests an explanation of a covered automated decision 1. Binding general data- protection law 2. The automated-decision provisionstookef- fect through the 2023 amendments and imple- menting decree 3. Does not create a gen- eral AI safety approval regime National Guide- linesforAI Ethics,2020 (Ministryof Scienceand ICT,Republic of Korea 2020) Government, developers, providers, users, andotherAI stakeholders 1. Organizes the guidance around human dignity, public good, and fitness for purpose 2. Sets ten requirements covering human rights, privacy,diversity, datamanagement, accountability,safety, transparency, solidarity, and related duties 1. Provides a nonbinding reference for the devel- opment and use of AI 1. Released by the Min- istry of Science and ICT in December 2020 2. Createsnodirect penaltyorapproval mechanism Table 13: South Korea AI Policy Overview PolicyInstru- ment Covered ActorsDocumented ProvisionsImplications for AI Sys- tems ImplementationStatus and Limitations InterimMea- suresforthe Management of GenerativeAI Services,2023 (Cyberspace Administration of China 2023) Providersof generativeAI services offered to the public in mainland China 1. Requires lawful training data and protection of intellectual property and personal information 2. Requires providers to address prohibited con- tent and protect users 3. Connectsservices withpublic-opinion attributesorsocial mobilization capacity to security assessment and algorithm filing rules 1. Applies to public gener- ative AI services rather than internal research and use 2. Requires measures on content,complaints, data, and service secu- rity 1. Effective from August 15, 2023 2. Binding administrative measure jointly issued by seven national au- thorities Measuresfor the Labeling of AI-Generated andSynthetic Content,2025 (Cyberspace Administration of China and others 2025) Generative AIproviders, content distribu- tion platforms, application distributors, and users 1. Requires explicit labels for specified generated or synthetic content 2. Requires implicit identi- fiers in file metadata 3. Sets duties for platforms that distribute or detect covered content 1. Adds provenance infor- mation to the production and distribution of syn- thetic content 2. Prohibitsmalicious removal, alteration, or concealment of required labels 1. Effective from Septem- ber 1, 2025 2. Complementsthe broader AI Plus de- velopmentinitiative (State Council of the People’s Republic of China 2025) Administrative Provisionson DeepSynthe- sisInternet Information Services,2022 (Cyberspace Administration of China 2022) Deepsynthe- sisservice providers, tech- nical supporters, and users 1. Requires identity verifi- cation and information security controls 2. Requires consent when editing biometric infor- mation such as faces or voices 3. Requires prominent la- bels for content that could confuse or mis- lead the public 1. Regulatescovered synthesisservices rather than banning the underlying techniques 2. Connects high-risk ser- vices to filing and secu- rity assessment duties 1. Effective from January 10, 2023 2. Binding administrative provisions Provisionson theAdmin- istrationof Algorithmic Recommen- dationsin InternetInfor- mation Services (Cyberspace Administration of China and others 2021) Algorithmic recommenda- tionservice providersand users 1. Requires disclosure of basic service principles, purposes, and operating mechanisms 2. Givesusersanon- personalized option or a convenient way to dis- ablerecommendation services 3. Requiresfilingfor services with public- opinionattributesor socialmobilization capacity 1. Regulates personaliza- tion,ranking,search filtering,dispatching, and related recommen- dation techniques 2. Sets additional protec- tions for minors, older people, workers, and consumers 1. Effective from March 1, 2022 2. Binding rules jointly is- sued by four national au- thorities Continued on next page Table 14 – continued from previous page PolicyInstru- ment Covered ActorsDocumented ProvisionsImplications for AI Sys- tems ImplementationStatus and Limitations Cybersecurity Law,Data SecurityLaw, andPersonal Information Protection Law (Standing Com- mitteeofthe NationalPeo- ple’sCongress of China 2016, 2021a,b) Networkop- erators,data processors, personalin- formation processors, and affected people 1. Establishes network se- curity and critical infras- tructure duties 2. Requiresclassified dataprotectionand safeguards for important data 3. Sets lawful bases, indi- vidual rights, processor duties, and cross-border rules for personal infor- mation 1. Applies to AI develop- ers and providers when their network, data, or personalinformation processing falls within scope 2. Does not amount to asingleAI-specific statute 1. Binding national laws with distinct scopes and enforcementmecha- nisms 2. The Cybersecurity Law was amended in 2025 with effect from January 1, 2026 Regulations on NetworkData SecurityMan- agement,2024 (StateCouncil of the People’s Republicof China 2024b) Networkdata processors and othercovered organizations 1. Implements duties un- der the Cybersecurity Law, Data Security Law, and Personal Informa- tion Protection Law 2. Addressespersonal information,impor- tantdata,platform duties, and cross-border transfers 1. Applies to AI-related data processing when it meets the regulation’s scope 2. Does not classify or ap- prove AI models as such 1. Promulgatedon September 24, 2024 2. Effective from January 1, 2025 Table 14: China AI Policy Overview C Existing Policy Evaluation Framework Review This appendix records the frameworks that contributed directly to the design synthesis. It is a selected methodological comparison, not a systematic literature review. PaperSummaryLimitations UNESCOReadiness AssessmentMethod- ology(UNESCO 2023) The methodology supports a national, multi-stakeholder assessment of legal, so- cial, economic, scientific, educational, and technical conditions for ethical AI. It com- bines document review, stakeholder partici- pation, and recommendations for policy re- form and institutional capacity. It is a diagnostic country process rather than a fixed rubric for comparing the content of versioned policy portfolios. Global Index on Re- sponsibleAI,2026 (Adams et al. 2026) The index assesses 135 countries and ju- risdictions through 38 indicators that cover policy and implementation, civil-society engagement, enabling conditions, and doc- umented government uses. It combines policy content with imple- mentation evidence and country conditions, while the present framework isolates the documented design of a declared policy portfolio. AI Governance Interna- tionaL Evaluation In- dex, 2025 (Zeng et al. 2025) The index evaluates 40 countries through four pillars, 17 dimensions, and 43 indica- tors drawn from policy documents, gover- nance practices, research outputs, and risk exposure. Its unit is national governance capacity rather than the content and legal force of a versioned policy portfolio. Government AI Readi- ness Index, 2025 (Ox- ford Insights 2025) The index evaluates 195 governments on their capacity to use, enable, and govern AI for public benefit. Readiness includes institutions, infrastruc- ture, skills, and private-sector conditions that lie outside the present rubric. A Grading Rubric for AI Safety Frameworks (Alaga, Schuett, and Anderljung 2024) The paper grades company frontier AI safety frameworks with seven criteria, 21 indicators, and letter grades from A to F. It also proposes surveys, Delphi studies, and audits as application methods. The unit is a company’s voluntary safety framework rather than a national portfolio of laws, strategies, budgets, and guidance. EvaluatingAI Providers’Frontier SafetyFrameworks (Stelling et al. 2025) The study applies 65 weighted criteria across four risk-management dimensions to 12 company frameworks. It shows how a detailed rubric can expose vague or missing commitments. The criteria address catastrophic-risk man- agement by frontier AI providers rather than national governance across sectors and institutions. ArtificialIntelli- gence Regulation: A Meta-Frameworkfor Formulation and Gov- ernance (Almeida, dos Santos Jr., and Farias 2020) The meta-framework connects ethical prin- ciples, accountability, risk management, stakeholder roles, and feedback between technology and regulation. It organizes policy formation and gover- nance but does not provide an operational scoring protocol for cross-national compar- ison. Assessing Frameworks for Eliciting Privacy and Security Require- ments from Laws and Regulations (Olukoya 2022) The paper assesses methods for translating legal rules into privacy and security require- ments and examines their coverage beyond the GDPR. The unit is a software requirements method, so the analysis informs individual criteria but not the structure of a national policy score. An AI System Eval- uationFramework forAdvancingAI Safety:Terminology, Taxonomy,Lifecycle Mapping (Xia et al. 2024) The framework harmonizes evaluation ter- minology, identifies elements of an AI system, and maps evaluations to lifecycle stages and supply-chain roles. It evaluates AI systems and their environ- ments rather than national policy portfolios. Evaluation Framework to Guide Implementa- tion of AI Systems into HealthcareSettings (Reddy et al. 2021) TEHAI evaluates capability, utility, and adoption throughout the development and deployment of clinical AI systems. Its clinical and system-level unit is narrower than national policy, although its attention to implementation and stakeholders informs several rubric items. Continued on next page Table 15 – continued from previous page PaperSummaryLimitations From AI to Digital Transformation:The AI Readiness Frame- work(Holmstr ̈ om 2022) The framework assesses organizational AI readiness across technologies, activi- ties, boundaries, and goals with a self- assessment scored from 0 to 4. It was illustrated in one insurance organiza- tion and measures organizational readiness rather than policy design. Implicationsofthe Use of Artificial In- telligenceinPublic Governance(Zuider- wijk, Chen, and Salem 2021) The systematic review organizes reported opportunities and risks of AI in government and sets an empirical research agenda for public governance. It synthesizes research on government use of AI rather than evaluating the content of national AI policy. AI Governance and the Policymaking Process (Perry and Uuk 2019) The paper connects AI governance with policy-process research and explains how political feasibility, timing, sequencing, and administrative capacity affect policy devel- opment. It provides a conceptual agenda rather than a rubric or comparative measurement pro- cedure. Table 15: Literature Review of AI Policy Frameworks D Complete National AI Policy Evaluation Framework The weights below implement the default rule in Section 3.3. Each category receives 20%. Criteria are equally weighted within each category, and sub-criteria are equally weighted within each criterion. Category 1. AI Risk Governance and Safety. Weight 20% CriterionDefinitionWeight 1. Safety StandardsThe policy includes specific and adaptable safety standards for AI systems, particularly for high-risk applications and advanced or general-purpose AI risks. 20% 2. Red Teaming and Auditing The policy mandates or strongly encourages independent audits, red teaming, and other forms of external scrutiny to identify vulnerabilities and ensure compliance. 20% 3. Human Oversight and Interventions The policy addresses the need for meaningful human control over AI systems, including mechanisms for intervention, override, or deactivation. 20% 4. System Interfacing and Delimitation The policy considers measures for containing or isolating potentially harmful AI systems to prevent unintended consequences or widespread damage. 20% 5. Long-Term AI Governance and Infrastructure Resilience The policy demonstrates an understanding of the long-term issues associated with governing advanced AI and infrastructure development. 20% Table 16: AI Risk Governance and Safety Criteria (1) Safety Standards. Sub-criterionDefinitionIndicators 1.1. Specificity of Safety Frameworks or Standards Are the safety standards clearly defined with specific technical requirements? 1. No specific safety standards mentioned. 2. Vague safety standards with little technical detail, deferred to future development. 3. Some specific safety standards are mentioned, but lack in detail or coverage, with concrete technical requirements and processes outlined for future work. 4. Clearly defined safety standards with considerable technical detail. 5. Comprehensive and highly specific safety standards with detailed technical requirements. 1.2. Coverage of Advanced AI Risks Do the safety standards address risks from advanced or general-purpose AI systems regardless of the terminology used? 1. No standards address risks from advanced or general-purpose AI systems. 2. The policy recognizes these risks but gives no concrete standard. 3. A concrete standard covers some relevant risks or systems. 4. Clear standards cover the main risks and state when they apply. 5. Detailed and adaptable standards cover capability growth, monitoring, and review. Continued on next page Table 17 – continued from previous page Sub-criterionDefinitionIndicators 1.3. High-Risk Application Focus Are there specific standards for high-risk applications (e.g., autonomous weapons, critical infrastructure)? 1. No differentiation of standards based on risk level. 2. Vague acknowledgment of high-risk applications. 3. Some specific standards for high-risk applications. 4. Clear and stringent standards for high-risk applications. 5. Comprehensive standards for high-risk applications with detailed provisions for risk mitigation. Table 17: Safety Standards Sub-criteria and Indicators (2) Red Teaming and Auditing. Sub-criterionDefinitionIndicators 2.1. Mandate for Independent Audits Does the policy mandate independent audits for AI systems, particularly for high-risk applications? 1. No mention of independent audits. 2. Audits encouraged but not mandated. 3. Audits mandated for some high-risk applications. 4. Audits mandated for most high-risk applications. 5. Audits mandated for all high-risk applications and strongly encouraged for others. 2.2. Misuse and Dual-Use Prevention Does the policy address the potential for AI technologies and systems to be misused or used for harmful purposes, considering their dual-use nature and the need for security measures to prevent unauthorized access, modification, or malicious use? 1. No consideration of the potential for misuse, dual-use risks, or the need for security measures to prevent unauthorized use. 2. Misuse and dual-use risks are acknowledged but not assessed or addressed through any concrete measures. 3. Some assessment of potential misuse and dual-use risks is mentioned, and some general security measures are outlined. 4. Clear guidelines for assessing and addressing misuse and dual-use risks are provided, along with clear and specific security requirements. 5. A comprehensive framework is in place for managing misuse and dual-use risks, including detailed risk assessment processes, guidelines and restrictions related to dual-use technologies, and comprehensive security requirements. 2.3. Red Teaming Requirements Does the policy specify requirements or guidelines for red teaming exercises? 1. No mention of red teaming. 2. Red teaming encouraged but not defined. 3. Red teaming mentioned with some guidelines. 4. Red teaming requirements are clearly defined. 5. Comprehensive red teaming requirements, including specific methodologies and scenarios. Continued on next page Table 18 – continued from previous page Sub-criterionDefinitionIndicators 2.4. External Scrutiny Mechanisms Does the policy establish mechanisms for external scrutiny beyond audits and red teaming (e.g., public reporting, expert panels)? 1. No mechanisms for external scrutiny. 2. Limited opportunities for external input. 3. Some mechanisms for external scrutiny. 4. Well-defined mechanisms for external scrutiny. 5. Multiple, robust mechanisms for external scrutiny, ensuring transparency and accountability. Table 18: Red Teaming and Auditing Sub-criteria and Indicators (3) Human Oversight and Interventions. Sub-criterionDefinitionIndicators 3.1. Human-in-the- Loop Requirements Does the policy mandate human-in-the-loop systems for critical decision-making, especially in high-risk applications? 1. No requirements for human oversight. 2. Human oversight is encouraged but not mandated. 3. Human-in-the-loop required for some critical decisions. 4. Human-in-the-loop required for most critical decisions. 5. Human-in-the-loop mandated for all critical decisions in high-risk applications. 3.2. Meaningful Human Control Does the policy define what constitutes meaningful human control over AI systems and how it can be maintained? 1. No definition of meaningful human control. 2. Meaningful human control is mentioned but not clearly defined. 3. Some aspects of meaningful human control are addressed. 4. Clear definition of meaningful human control. 5. Comprehensive framework for ensuring meaningful human control, including provisions for human override and intervention. 3.3. Oversight of Autonomous Systems Does the policy address the specific challenges of overseeing increasingly autonomous AI systems? 1. No consideration of oversight for autonomous systems. 2. Oversight of autonomous systems is mentioned but not addressed in detail. 3. Some guidelines for overseeing autonomous systems. 4. Clear guidelines and requirements for overseeing autonomous systems. 5. Comprehensive framework for overseeing autonomous systems, including provisions for monitoring, auditing, and accountability. Continued on next page Table 19 – continued from previous page Sub-criterionDefinitionIndicators 3.4. Override and Deactivation Mechanisms Does the policy specify clear mechanisms for human operators to intervene in AI system operations, including overriding decisions and safely deactivating or shutting down systems in case of emergency, malfunction, or other critical situations? 1. The policy does not mention override, intervention, or deactivation mechanisms. 2. Mechanisms for override or deactivation are alluded to or mentioned as possible but are not clearly defined or detailed. 3. Mechanisms for human override of AI decisions or system deactivation are defined for some critical scenarios, but not comprehensively for both, or are not easily accessible. 4. Clear mechanisms for both human override of AI decisions and system deactivation (including emergency shutdown) are defined for most critical scenarios, with reasonable accessibility. 5. Comprehensive, easily accessible, and robust mechanisms for both human override of AI decisions and system deactivation (including emergency shutdown procedures) are clearly defined, well-documented, and readily available for all relevant scenarios. Table 19: Human Oversight and Interventions Sub-criteria and Indicators (4) System Interfacing and Delimitation. Sub-criterionDefinitionIndicators 4.1. Research and Development Guidelines Does the policy provide guidelines for research and development of containment or isolation techniques for advanced AI? 1. No mention of containment or isolation research. 2. Containment research is encouraged but not prioritized. 3. Some guidelines for containment research. 4. Clear guidelines for containment research and development. 5. Comprehensive guidelines and funding for containment research, recognizing its importance. 4.2. Deployment Restrictions Does the policy restrict the deployment of AI systems that cannot be adequately contained or isolated? 1. No restrictions on deployment based on containment. 2. Vague restrictions with unclear criteria. 3. Some restrictions based on containment feasibility. 4. Clear restrictions based on containment feasibility for high-risk applications. 5. Comprehensive restrictions for any AI system that cannot be adequately contained or isolated. 4.3. Emergency Protocols Does the policy include provisions for emergency protocols in case of containment or isolation failures? 1. No mention of emergency protocols. 2. Emergency protocols are underdeveloped. 3. Some emergency protocols are defined. 4. Clear emergency protocols are established. 5. Comprehensive emergency protocols, including backup containment and mitigation strategies. Table 20: System Interfacing and Delimitation Sub-criteria and Indicators (5) Long-Term AI Governance and Infrastructure Resilience. Sub-criterionDefinitionIndicators 5.1. Adaptive Governance Does the policy include mechanisms for adapting governance structures and regulations as AI technology, particularly AGI, advances? 1. No mechanisms for adaptation. 2. Adaptation is possible but not clearly defined. 3. Some mechanisms for adaptation based on new information. 4. Clear process for reviewing and updating policies in response to AGI advancements. 5. Flexible and adaptive governance framework with built-in mechanisms for continuous learning and adjustment. 5.2. AI Supply Chain Security and Resilience Does the policy address security risks within the AI supply chain (hardware, software, data, vendors), promotes resilience against disruptions or compromises, and includes strategic measures to foster domestic capabilities, aiming to reduce critical dependencies and enhance national self-sufficiency in key AI technologies. 1. No mention of AI supply chain risks, resilience, or national dependencies. 2. Vague acknowledgement of supply chain risks with no specific requirements or strategies for domestic capacity. 3. Identifies some AI supply chain risks and recommends basic security practices or domestic sourcing preference, but lacks concrete strategic plans, dedicated funding, or specific incentives for AI self-sufficiency. 4. Requires specific risk management practices for critical AI supply chain elements and includes targeted initiatives with dedicated funding or strong incentives to bolster domestic capacity and reduce dependencies in key AI technology areas. 5. Mandates a comprehensive national strategy for AI supply chain security and resilience, integrating robust risk management, resilience plans against disruptions, significant investment in domestic self-sufficiency, and clear goals to reduce critical foreign dependencies. Table 21: Long-Term AI Governance and Infrastructure Resilience Sub-criteria and Indicators Category 2. Effectiveness and Feasibility. Weight 20% CriterionDefinitionWeight 6. Institutional Capacity and Authority The policy establishes appropriately empowered, independent, resourced, and coordinated institutions to implement, oversee, and enforce it. 16.67% 7. Clarity and SpecificityThe policy provisions are clearly defined, specific, and actionable, avoiding ambiguity and leaving little room for misinterpretation. 16.67% 8. EnforceabilityThe policy includes mechanisms for monitoring compliance, detecting violations, and enforcing its provisions, including appropriate penalties for non-compliance. 16.67% 9. MeasurabilityThe policy objectives and intended outputs are measurable, allowing later assessment of implementation and results. 16.67% 10. Resource Requirements The resources required for implementing and enforcing the policy are realistic, feasible, and proportionate to the policy’s objectives. 16.67% 11. Incentive AlignmentThe policy aligns incentives for developers, deployers, users, and other affected actors to promote responsible AI development and deployment. 16.67% Table 22: Effectiveness and Feasibility Criteria (6) Institutional Capacity and Authority. Sub-criterionDefinitionIndicators 6.1. Dedicated Oversight Body Existence and specific mandate of a body (or bodies) primarily responsible for AI policy implementation, oversight, and enforcement. 1. No specific body identified or proposed for AI oversight. Responsibility is unassigned. 2. Mentions the need for oversight but relies on existing bodies without clearly expanding their mandate or resources for AI. 3. Designates an existing body or proposes a new body but with an unclear, limited, or overlapping mandate specific to AI governance. 4. Clearly designates or establishes a specific body with a defined mandate covering key aspects of the AI policy implementation and oversight. 5. Establishes or clearly empowers a dedicated body with a comprehensive, clearly defined, and primary mandate covering all major policy areas, including monitoring, guidance, and coordinating enforcement. 6.2. Independence and Authority Level of operational independence from political or corporate influence and the legal authority granted to the oversight body. 1. The oversight body (if any) has no specified independence or distinct legal authority for AI matters. 2. Vague statements about needing authority or impartiality, but no structural guarantees or specific powers defined. 3. Some measures for independence are outlined (e.g., reporting structure), but legal authority for key actions (e.g., investigation, binding guidance) is limited or unclear. 4. Clear provisions ensuring significant operational independence and defined legal authority for key functions (e.g., investigation, issuing fines, setting standards within delegated powers). 5. Strong structural/statutory independence guaranteed, coupled with comprehensive legal authority (e.g., independent rule-making within its mandate, broad investigative powers, direct enforcement capabilities). 6.3. Resources and Expertise Adequacy of dedicated funding, staffing, and necessary technical, legal, and ethical expertise within the oversight body. 1. No mention of specific resource allocation or expertise requirements for the oversight body. 2. Acknowledges the need for resources/expertise but provides no concrete commitments or plans. 3. Some resource allocation is mentioned but does not match the mandate. The policy refers generally to a need for diverse expertise. 4. Specifies dedicated funding streams or significant budget allocation and mandates the inclusion of specific technical, ethical, and legal expertise. 5. Guarantees substantial, ring-fenced, long-term funding, mandates specific high-level expertise across relevant domains, and includes plans for talent recruitment, retention, and training. Continued on next page Table 23 – continued from previous page Sub-criterionDefinitionIndicators 6.4. Interagency Coordination Mechanisms Existence and operation of formal coordination and information-sharing procedures among government bodies involved in AI governance. 1. No mention of coordination between different government agencies on AI. 2. Vague mention of the need for collaboration without establishing specific mechanisms. 3. Establishes informal or ad-hoc coordination mechanisms (e.g., occasional meetings). 4. Establishes formal coordination structures (e.g., standing inter-agency committee or task force) with defined roles and responsibilities for AI policy. 5. Establishes robust, permanent coordination mechanisms with clear mandates, leadership, regular reporting, dedicated resources, and clear protocols for information sharing and joint action on AI governance. Table 23: Institutional Capacity and Authority Sub-criteria and Indicators (7) Clarity and Specificity. Sub-criterionDefinitionIndicators 7.1. Implementation Plan Clarity, feasibility, and level of detail provided in the policy regarding its implementation plan, including phases, timelines, milestones, and responsible parties. 1. No implementation plan or timeline outlined in the policy. 2. Vague statements about future implementation or gradual rollout without specifics. 3. Outlines some implementation steps or phases but lacks clear timelines, measurable milestones, or specific assignment of responsibilities. 4. Provides a clear implementation plan with distinct phases, key milestones, assigned responsibilities for major actions, and indicative timelines. 5. Provides a detailed, phased implementation plan with specific and measurable milestones, concrete timelines, clearly assigned responsibilities across agencies, and links to resource allocation. Table 24: Clarity and Specificity Sub-criteria and Indicators (8) Enforceability. Sub-criterionDefinitionIndicators 8.1. Technical Feasibility Are the technical requirements of the policy feasible to implement with current or near-future technology? 1. Policy requirements are largely unimplementable with current technology. 2. Most requirements are technically challenging or impractical. 3. A notable number of requirements are technically feasible, but others are challenging. 4. Most requirements are technically feasible. 5. All requirements are technically feasible with current or near-future technology. Continued on next page Table 25 – continued from previous page Sub-criterionDefinitionIndicators 8.2. Compliance Monitoring Does the policy establish clear mechanisms for monitoring compliance? 1. No mechanisms for monitoring compliance. 2. Monitoring mechanisms are vaguely defined or inadequate. 3. Some mechanisms for monitoring compliance are outlined. 4. Well-defined mechanisms for monitoring compliance. 5. Comprehensive and robust mechanisms for monitoring compliance, including regular audits and reporting. 8.3. Enforcement Mechanisms Does the policy specify enforcement mechanisms and penalties for non-compliance? 1. No enforcement mechanisms or penalties specified. 2. Enforcement mechanisms are weak or unclear. 3. Some enforcement mechanisms and penalties are defined. 4. Clear enforcement mechanisms and appropriate penalties are specified. 5. Comprehensive enforcement mechanisms with a range of penalties, including provisions for addressing serious violations. 8.4. Legal Status and Force Assessment of the policy document’s formal legal standing and its binding nature on relevant actors. 1. Policy is purely informational, aspirational, or strategic (e.g., national strategy document, white paper) with no legal obligations. 2. Policy takes the form of voluntary guidelines, recommendations, or a code of conduct with no direct legal enforceability. 3. Policy uses soft regulatory tools (e.g., mandatory for government procurement only, relies on certification schemes) or has limited binding applicability (e.g., specific narrow sector). 4. Policy is a binding regulation or executive order creating clear legal obligations for defined actors/sectors, with established legal consequences for non-compliance. 5. Policy is enshrined in comprehensive, legally binding primary legislation (Statute/Act) with broad applicability across relevant sectors and strong, clearly defined enforcement powers and penalties. Table 25: Enforceability Sub-criteria and Indicators (9) Measurability. Sub-criterionDefinitionIndicators 9.1. Quantifiable Objectives Does the policy define quantifiable objectives that can be used to measure its success? 1. No quantifiable objectives defined. 2. Objectives are vaguely defined and difficult to measure. 3. Some objectives are quantifiable, but others are not. 4. Most objectives are quantifiable. 5. All key objectives are clearly defined and quantifiable. Continued on next page Table 26 – continued from previous page Sub-criterionDefinitionIndicators 9.2. Data Collection and Reporting Does the policy include provisions for collecting and reporting data relevant to the policy’s objectives and outcomes? 1. No provisions for data collection or reporting. 2. Data collection and reporting are vaguely defined. 3. Some data collection and reporting requirements are outlined. 4. Clear data collection and reporting requirements are specified. 5. Comprehensive data collection and reporting requirements, including provisions for data quality and transparency. 9.3. Performance Metrics Does the policy define specific performance metrics to assess the effectiveness of AI systems in meeting the policy’s objectives? 1. No performance metrics defined. 2. Performance metrics are vaguely defined or inadequate. 3. Some performance metrics are defined. 4. Clear and relevant performance metrics are specified. 5. Comprehensive set of performance metrics aligned with policy objectives. Table 26: Measurability Sub-criteria and Indicators (10) Resource Requirements. Sub-criterionDefinitionIndicators 10.1. Financial Resources Does the policy consider the financial resources required for implementation and enforcement? 1. No consideration of financial resource requirements. 2. Financial resource requirements are underestimated or unrealistic. 3. Some consideration of financial resource requirements. 4. Realistic assessment of financial resource requirements. 5. Comprehensive analysis of financial resource requirements, including provisions for funding and cost-effectiveness. 10.2. Human Resources Does the policy consider the human resources (e.g., expertise, personnel) required for implementation and enforcement? 1. No consideration of human resource requirements. 2. Human resource requirements are underestimated or unrealistic. 3. Some consideration of human resource requirements. 4. Realistic assessment of human resource requirements. 5. Comprehensive analysis of human resource requirements, including provisions for training and capacity building. Continued on next page Table 27 – continued from previous page Sub-criterionDefinitionIndicators 10.3. Technological Resources Does the policy consider the technological resources (e.g., infrastructure, tools, data) required for implementation and enforcement? 1. No consideration of technological resource requirements. 2. Technological resource requirements are underestimated or unrealistic. 3. Some consideration of technological resource requirements. 4. Realistic assessment of technological resource requirements. 5. Comprehensive analysis of technological resource requirements, including provisions for infrastructure and tools. Table 27: Resource Requirements Sub-criteria and Indicators (11) Incentive Alignment. Sub-criterionDefinitionIndicators 11.1. Positive Incentives Does the policy provide positive incentives (e.g., rewards, recognition) for responsible AI practices? 1. No positive incentives provided. 2. Positive incentives are vaguely defined or weak. 3. Some positive incentives are provided. 4. Clear and meaningful positive incentives are offered. 5. Comprehensive set of positive incentives that effectively promote responsible AI practices. 11.2. Negative Incentives Does the policy include negative incentives (e.g., penalties, sanctions) for irresponsible AI practices? 1. No negative incentives specified. 2. Negative incentives are weak or unclear. 3. Some negative incentives are defined. 4. Clear and proportionate negative incentives are specified. 5. Comprehensive set of negative incentives that effectively deter irresponsible AI practices. 11.3. Incentive Compatibility Are the incentives designed to be compatible with each other and with the overall objectives of the policy? 1. Incentives are contradictory or misaligned. 2. Incentives are weakly aligned with each other and with policy objectives. 3. Some incentives are aligned, but others are not. 4. Most incentives are aligned with each other and with policy objectives. 5. Incentives are fully compatible with each other and strongly aligned with the overall objectives of the policy. Table 28: Incentive Alignment Sub-criteria and Indicators Category 3. Comprehensiveness and Scope. Weight 20% CriterionDefinitionWeight 12. Coverage of AI Lifecycle The policy addresses the full lifecycle of AI systems, from initial design and data collection to development, deployment, monitoring, and eventual decommissioning. 20% 13. Range of Risks Addressed The policy covers a wide range of risks associated with AI systems, including technical, ethical, societal, and governance risks. 20% Continued on next page Table 29 – continued from previous page CriterionDefinitionWeight 14. Stakeholder InclusionThe policy considers the perspectives and roles of developers, deployers, users, regulators, and affected communities. 20% 15. Adaptability to Technological Advancements The policy includes mechanisms for adapting to new and emerging AI technologies, including provisions for ongoing research, monitoring, and updates. 20% 16. International Harmonization The policy aligns with or considers international standards, practices, and agreements related to AI governance, promoting consistency and cooperation across borders. 20% Table 29: Comprehensiveness and Scope Criteria (12) Coverage of AI Lifecycle. Sub-criterionDefinitionIndicators 12.1. Design PhaseDoes the policy include provisions for responsible AI design principles and practices? 1. No consideration of the design phase. 2. Design phase is mentioned but not addressed in detail. 3. Some guidelines for responsible AI design. 4. Clear guidelines and requirements for responsible AI design. 5. Comprehensive framework for responsible AI design, including ethical considerations and risk assessment. 12.2. Data Collection and Preprocessing Does the policy address data collection, preprocessing, and governance practices? 1. No consideration of data practices. 2. Data practices are mentioned but not addressed in detail. 3. Some guidelines for responsible data practices. 4. Clear guidelines and requirements for data collection, preprocessing, and governance. 5. Comprehensive framework for responsible data practices, including provisions for data quality, fairness, and privacy. 12.3. Development and Testing Does the policy include provisions for responsible AI development and testing methodologies? 1. No consideration of development and testing. 2. Development and testing are mentioned but not addressed in detail. 3. Some guidelines for responsible development and testing. 4. Clear guidelines and requirements for development and testing, including validation and verification. 5. Comprehensive framework for responsible development and testing, including provisions for robustness, security, and explainability. Continued on next page Table 30 – continued from previous page Sub-criterionDefinitionIndicators 12.4. Deployment and Monitoring Does the policy address the deployment and ongoing monitoring of AI systems? 1. No consideration of deployment and monitoring. 2. Deployment and monitoring are mentioned but not addressed in detail. 3. Some guidelines for responsible deployment and monitoring. 4. Clear guidelines and requirements for deployment and monitoring, including performance tracking and risk assessment. 5. Comprehensive framework for responsible deployment and monitoring, including provisions for continuous evaluation and adaptation. 12.5. Decommissioning Does the policy address the eventual decommissioning or retirement of AI systems? 1. No consideration of decommissioning. 2. Decommissioning is mentioned but not addressed in detail. 3. Some guidelines for responsible decommissioning. 4. Clear guidelines and requirements for decommissioning. 5. Comprehensive framework for responsible decommissioning, including provisions for data retention, model updates, and system retirement. Table 30: Coverage of AI Lifecycle Sub-criteria and Indicators (13) Range of Risks Addressed. Sub-criterionDefinitionIndicators 13.1. Technical Risks Does the policy address technical risks such as safety, security, robustness, reliability, and integration with existing non-AI systems and infrastructure? 1. No consideration of technical risks. 2. Technical risks are mentioned but not addressed in detail. 3. Some guidelines for addressing technical risks. 4. Clear guidelines and requirements for mitigating technical risks. 5. Comprehensive framework for managing technical risks, including specific standards and best practices. 13.2. Ethical RisksDoes the policy address ethical risks such as bias, fairness, transparency, accountability, and privacy? 1. No consideration of ethical risks. 2. Ethical risks are mentioned but not addressed in detail. 3. Some guidelines for addressing ethical risks. 4. Clear guidelines and requirements for mitigating ethical risks. 5. Comprehensive framework for managing ethical risks, including provisions for ethical audits and impact assessments. Continued on next page Table 31 – continued from previous page Sub-criterionDefinitionIndicators 13.3. Societal Risks Does the policy address societal risks such as job displacement, inequality, social manipulation, and erosion of trust? 1. No consideration of societal risks. 2. Societal risks are mentioned but not addressed in detail. 3. Some guidelines for addressing societal risks. 4. Clear guidelines and requirements for mitigating societal risks. 5. Comprehensive framework for managing societal risks, including provisions for public engagement and social impact assessments. 13.4. Governance Risks Does the policy address governance risks such as lack of accountability, regulatory gaps, and challenges to international cooperation? 1. No consideration of governance risks. 2. Governance risks are mentioned but not addressed in detail. 3. Some guidelines for addressing governance risks. 4. Clear guidelines and requirements for mitigating governance risks. 5. Comprehensive framework for managing governance risks, including provisions for regulatory oversight and international collaboration. Table 31: Range of Risks Addressed Sub-criteria and Indicators (14) Stakeholder Inclusion. Sub-criterionDefinitionIndicators 14.1. Stakeholder Identification Does the policy clearly identify all relevant stakeholders affected by AI systems? 1. No stakeholders are identified. 2. Some stakeholders are identified, but with significant gaps. 3. Most relevant stakeholders are identified. 4. All relevant stakeholders are clearly identified. 5. Comprehensive stakeholder identification, including marginalized and underrepresented groups. 14.2. Stakeholder Engagement Does the policy include mechanisms for engaging stakeholders in the policy development and implementation process? 1. No mechanisms for stakeholder engagement. 2. Limited opportunities for stakeholder input. 3. Some mechanisms for stakeholder engagement. 4. Well-defined mechanisms for stakeholder engagement. 5. Multiple, robust mechanisms for stakeholder engagement, ensuring diverse perspectives are considered. 14.3. Stakeholder Roles and Responsibilities Does the policy clearly define the roles and responsibilities of different stakeholders in relation to AI safety and governance? 1. Roles and responsibilities are not defined. 2. Roles and responsibilities are vaguely mentioned. 3. Some roles and responsibilities are defined, but gaps remain. 4. Most roles and responsibilities are clearly defined. 5. Comprehensive definition of roles and responsibilities for all relevant stakeholders, including provisions for accountability and redress. Table 32: Stakeholder Inclusion Sub-criteria and Indicators (15) Adaptability to Technological Advancements. Sub-criterionDefinitionIndicators 15.1. Technology Monitoring Does the policy establish mechanisms for monitoring advancements in AI technology and their potential implications? 1. No provisions for technology monitoring. 2. Technology monitoring is mentioned but not clearly defined. 3. Some mechanisms for technology monitoring are outlined. 4. Clear mechanisms for monitoring technological advancements. 5. Comprehensive framework for monitoring technological advancements, including early warning systems for emerging risks. 15.2. Policy Review and Update Does the policy include a process for regularly reviewing and updating its provisions in response to new technological developments? 1. No process for policy review or update. 2. Policy review is possible but not clearly defined or mandated. 3. Some provisions for policy review and update. 4. Clear process for regular policy review and update. 5. Comprehensive process for policy review and update, including triggers for review based on specific technological advancements. 15.3. Flexibility and Adaptability Is the policy framework designed to be flexible and adaptable to accommodate new AI technologies and applications? 1. Policy framework is rigid and inflexible. 2. Limited flexibility to accommodate new technologies. 3. Some flexibility but may not be sufficient for all new technologies. 4. Policy framework is designed to be adaptable to most new technologies. 5. Highly flexible and adaptable framework that can accommodate a wide range of emerging AI technologies, applications, and sudden changes to the AI landscape of a nation. Continued on next page Table 33 – continued from previous page Sub-criterionDefinitionIndicators 15.4. Incident reporting, review and global Monitoring Existence of mechanisms for reporting AI incidents/failures or events and processes for analyzing these incidents to adapt policy and best practices. 1. No formal mechanisms for internal incident reporting/review or systematic global AI monitoring. 2. Informal or ad hoc approaches to incident reporting or global monitoring with no defined structure, analysis, or feedback. 3. Establishes formal but limited/voluntary internal incident reporting or assigns responsibility for basic global monitoring, but lacks systematic analysis and policy feedback integration for either. 4. Formal internal incident reporting/review process (may be limited in scope) and a structured process for regular global AI monitoring and reporting to inform policy reviews. 5. Comprehensive, mandatory internal incident reporting, review, and learning system and robust, continuous global AI monitoring with dedicated analysis and a formal, rapid feedback loop for policy/strategy adaptation. Table 33: Adaptability to Technological Advancements Sub-criteria and Indicators (16) International Harmonization. Sub-criterionDefinitionIndicators 16.1. Alignment with International Standards Does the policy align with relevant international standards and guidelines for AI governance? 1. No consideration of international standards. 2. Limited alignment with international standards. 3. Some alignment with international standards. 4. Strong alignment with international standards. 5. Comprehensive alignment with international standards, including active participation in their development. 16.2. Cross-Border Data Flows Does the policy address the challenges of cross-border data flows and data governance in the context of AI systems? 1. No consideration of cross-border data flows. 2. Cross-border data flows are mentioned but not addressed in detail. 3. Some guidelines for managing cross-border data flows. 4. Clear guidelines and requirements for managing cross-border data flows. 5. Comprehensive framework for managing cross-border data flows, including provisions for data localization, data sovereignty, and international data sharing agreements. Continued on next page Table 34 – continued from previous page Sub-criterionDefinitionIndicators 16.3. International Cooperation Does the policy promote international cooperation in AI governance, including mechanisms for information sharing, joint research, and coordinated policy development? 1. No provisions for international cooperation. 2. International cooperation is encouraged but not prioritized. 3. Some mechanisms for international cooperation are outlined. 4. Clear commitments to international cooperation, including specific initiatives and partnerships. 5. Comprehensive framework for international cooperation, including mechanisms for joint research, information sharing, and coordinated policy development. Table 34: International Harmonization Sub-criteria and Indicators Category 4. User Rights, Protection and Agency. Weight 20% CriterionDefinitionWeight 17. Fairness and Non-discrimination The policy addresses bias, fairness, and non-discrimination in AI systems, including provisions for identifying, mitigating, and monitoring bias. 25% 18. Transparency and Explainability The policy promotes transparency and explainability in AI decision-making, enabling affected people to understand relevant outputs or decisions. 25% 19. Privacy and Data Protection The policy includes provisions for protecting privacy and securing personal data used in AI systems under the applicable data protection rules. 25% 20. AccountabilityThe policy establishes responsibility for the development, deployment, and use of AI systems, including mechanisms for liability and redress when appropriate. 25% Table 35: User Rights, Protection and Agency Criteria (17) Fairness and Non-discrimination. Sub-criterionDefinitionIndicators 17.1. Bias Detection and Mitigation Does the policy require or encourage the use of methods for detecting and mitigating bias in AI systems? 1. No consideration of bias. 2. Bias is mentioned but not addressed in detail. 3. Some guidelines for bias detection and mitigation. 4. Clear guidelines and requirements for bias detection and mitigation. 5. Comprehensive framework for bias detection and mitigation, including specific methodologies and metrics. 17.2. Protected Groups Does the policy explicitly address the potential impact of AI systems on protected groups (e.g., based on race, gender, religion)? 1. No consideration of protected groups. 2. Protected groups are mentioned but not addressed in detail. 3. Some guidelines for protecting the rights of protected groups. 4. Clear guidelines and requirements for protecting the rights of protected groups. 5. Comprehensive framework for protecting the rights of protected groups, including provisions for auditing and redress. Table 36: Fairness and Non-discrimination Sub-criteria and Indicators (18) Transparency and Explainability. Sub-criterionDefinitionIndicators 18.1. Transparency Requirements Does the policy specify transparency requirements for AI systems, including documentation of data, algorithms, and decision-making processes? 1. No transparency requirements. 2. Transparency is encouraged but not clearly defined. 3. Some transparency requirements are outlined. 4. Clear and specific transparency requirements are specified. 5. Comprehensive transparency requirements, including provisions for data provenance, model documentation, and decision-making logic. 18.2. Explainability Methods Does the policy encourage or mandate the use of explainability methods to make AI systems more understandable to humans? 1. No consideration of explainability. 2. Explainability is mentioned but not addressed in detail. 3. Some guidelines for using explainability methods. 4. Clear guidelines and recommendations for using explainability methods. 5. Comprehensive framework for explainability, including provisions for different types of explanations (e.g., local, global) and different stakeholder needs. 18.3. Interpretability for Non-Experts Does the policy consider the need for interpretability by non-experts, including lay users and affected communities? 1. No consideration of interpretability for non-experts. 2. Interpretability for non-experts is mentioned but not prioritized. 3. Some guidelines for making AI systems understandable to non-experts. 4. Clear guidelines and requirements for interpretability by non-experts. 5. Comprehensive framework for interpretability by non-experts, including provisions for user-friendly explanations and interfaces. Table 37: Transparency and Explainability Sub-criteria and Indicators (19) Privacy and Data Protection. Sub-criterionDefinitionIndicators 19.1. Data Minimization and Purpose Limitation Does the policy promote data minimization principles, limiting the collection and use of personal data to what is strictly necessary for the intended purpose? 1. No consideration of data minimization. 2. Data minimization is encouraged but not clearly defined. 3. Some guidelines for data minimization. 4. Clear requirements for data minimization and purpose limitation. 5. Comprehensive framework for data minimization and purpose limitation, including provisions for data deletion and retention policies. Continued on next page Table 38 – continued from previous page Sub-criterionDefinitionIndicators 19.2. Data SecurityDoes the policy mandate appropriate security measures to protect personal data from unauthorized access, use, or disclosure? 1. No specific security measures mentioned. 2. Security measures are encouraged but not defined. 3. Some security measures are outlined. 4. Clear and specific security requirements are specified. 5. Comprehensive security requirements, including encryption, access controls, and regular security audits. 19.3. Consent and Individual Rights Does the policy ensure that individuals have meaningful control over their personal data, including the right to access, rectify, erase, and restrict processing? 1. No consideration of individual rights. 2. Individual rights are mentioned but not fully addressed. 3. Some provisions for individual rights. 4. Clear mechanisms for individuals to exercise their rights. 5. Comprehensive framework for individual rights, including provisions for informed consent, data portability, and the right to be forgotten. Table 38: Privacy and Data Protection Sub-criteria and Indicators (20) Accountability. Sub-criterionDefinitionIndicators 20.1. Responsibility Allocation Does the policy clearly define the responsibilities of different actors in the AI lifecycle (e.g., developers, deployers, users)? 1. Responsibilities are not defined. 2. Responsibilities are vaguely mentioned. 3. Some responsibilities are defined, but gaps remain. 4. Most responsibilities are clearly defined. 5. Comprehensive definition of responsibilities for all relevant actors. 20.2. Liability Framework Does the policy establish a liability framework for determining who is liable when AI systems cause harm or errors? 1. No liability framework established. 2. Liability framework is unclear or inadequate. 3. Some aspects of liability are addressed. 4. Clear liability framework for most scenarios. 5. Comprehensive liability framework that addresses various types of harm and different levels of AI autonomy. 20.3. Redress Mechanisms Does the policy provide mechanisms for individuals or groups to seek redress for harm caused by AI systems? 1. No redress mechanisms provided. 2. Redress mechanisms are limited or difficult to access. 3. Some redress mechanisms are available. 4. Clear and accessible redress mechanisms are established. 5. Comprehensive redress mechanisms, including provisions for compensation, remediation, and appeal. Table 39: Accountability Sub-criteria and Indicators Category 5. Socioeconomic Impact and Innovation. Weight 20% CriterionDefinitionWeight 21. Societal BenefitThe policy promotes the development and use of AI systems that provide public benefits in areas such as healthcare, education, environmental sustainability, and economic activity. 20% 22. InnovationThe policy fosters responsible AI research, development, and deployment while addressing relevant risks. 20% 23. Economic GrowthThe policy supports economic development through AI adoption while addressing challenges such as job displacement and market disruption. 20% 24. Social EquityThe policy addresses how AI benefits and harms are distributed, including effects on marginalized or vulnerable groups. 20% 25. Public TrustThe policy supports warranted public trust through transparency, accountability, public education, and opportunities for participation. 20% Table 40: Socioeconomic Impact and Innovation Criteria (21) Societal Benefit. Sub-criterionDefinitionIndicators 21.1. Identification of Beneficial Applications Does the policy identify specific areas where AI can provide significant societal benefits? 1. No identification of beneficial applications. 2. Beneficial applications are mentioned but not prioritized. 3. Some areas of societal benefit are identified. 4. Clear identification of key areas where AI can provide significant benefits. 5. Comprehensive analysis of potential societal benefits across various sectors. 21.2. Incentives for Beneficial AI Does the policy provide incentives (e.g., funding, grants, recognition) for the development and deployment of AI systems that address societal needs? 1. No incentives for beneficial AI. 2. Incentives are weak or unclear. 3. Some incentives for beneficial AI are provided. 4. Clear and meaningful incentives for beneficial AI. 5. Comprehensive set of incentives that effectively promote the development and deployment of beneficial AI. 21.3. Measurement of Societal Impact Does the policy include mechanisms for measuring and evaluating the societal impact of AI systems? 1. No mechanisms for measuring societal impact. 2. Societal impact is mentioned but not measured. 3. Some metrics for measuring societal impact are proposed. 4. Clear metrics for measuring societal impact. 5. Comprehensive framework for measuring and evaluating the societal impact of AI, including both positive and negative impacts. Table 41: Societal Benefit Sub-criteria and Indicators (22) Innovation. Sub-criterionDefinitionIndicators 22.1. Support for Research and Development Does the policy provide support for research and development in AI, including funding, infrastructure, and talent development? 1. No support for AI research and development. 2. Limited support for AI research and development. 3. Some funding and support for AI research and development. 4. Significant funding and support for AI research and development. 5. Comprehensive support for AI research and development, including long-term investments and strategic initiatives. 22.2. Regulatory Sandboxes and Testbeds Does the policy promote the use of regulatory sandboxes or testbeds to experiment with new AI technologies and applications in a controlled environment? 1. No mention of regulatory sandboxes or testbeds. 2. Sandboxes or testbeds are mentioned but not clearly defined. 3. Some provisions for regulatory sandboxes or testbeds. 4. Clear guidelines and procedures for establishing and operating regulatory sandboxes or testbeds. 5. Comprehensive framework for regulatory sandboxes or testbeds, including provisions for data sharing, risk assessment, and stakeholder engagement. 22.3. Innovation- Friendly Regulation Is the policy framework designed to be flexible and adaptable to accommodate new AI innovations without stifling their development? 1. Policy framework is rigid and inflexible, hindering innovation. 2. Limited flexibility to accommodate new innovations. 3. Some flexibility but may not be sufficient for all new innovations. 4. Policy framework is designed to be adaptable to most new innovations. 5. Highly flexible and adaptable framework that encourages responsible innovation while mitigating risks. Table 42: Innovation Sub-criteria and Indicators (23) Economic Growth. Sub-criterionDefinitionIndicators 23.1. Investment in AI Ecosystem Does the policy promote investment in the AI ecosystem, including startups, businesses, and research institutions? 1. No provisions for promoting investment in the AI ecosystem. 2. Investment is encouraged but not prioritized. 3. Some incentives for investing in the AI ecosystem. 4. Clear incentives and support for investment in the AI ecosystem. 5. Comprehensive strategy for promoting investment in the AI ecosystem, including tax breaks, grants, and venture capital initiatives. Continued on next page Table 43 – continued from previous page Sub-criterionDefinitionIndicators 23.2. Workforce Development Does the policy address the need for workforce development and retraining to prepare workers for the changing job market due to AI adoption? 1. No consideration of workforce development. 2. Workforce development is mentioned but not addressed in detail. 3. Some guidelines for workforce development and retraining. 4. Clear guidelines and programs for workforce development and retraining. 5. Comprehensive framework for workforce development, including provisions for education, training, and job placement. 23.3. Market Competitiveness Does the policy promote fair competition in the AI market, preventing monopolies and ensuring a level playing field for businesses of all sizes? 1. No consideration of market competitiveness. 2. Market competitiveness is mentioned but not addressed in detail. 3. Some guidelines for promoting fair competition. 4. Clear guidelines and regulations to prevent monopolies and promote fair competition. 5. Comprehensive framework for ensuring market competitiveness, including antitrust measures and support for small and medium-sized enterprises. Table 43: Economic Growth Sub-criteria and Indicators (24) Social Equity. Sub-criterionDefinitionIndicators 24.1. Access to AI Benefits Does the policy promote equitable access to the benefits of AI technologies, regardless of socioeconomic status, geographic location, or other factors? 1. No consideration of equitable access to AI benefits. 2. Equitable access is mentioned but not prioritized. 3. Some initiatives to promote equitable access. 4. Clear guidelines and programs to ensure equitable access. 5. Comprehensive framework for promoting equitable access to AI benefits, including targeted interventions for marginalized communities. 24.2. Social Safety Nets Does the policy consider the need for social safety nets like Universal Basic Income (UBI) to support individuals and communities affected by job displacement or other negative consequences of AI adoption? 1. No consideration of social safety nets. 2. Social safety nets are mentioned but not addressed in detail. 3. Some provisions for social safety nets. 4. Clear guidelines and programs for social safety nets. 5. Comprehensive framework for social safety nets, including provisions for retraining, job placement, and income support (UBI). Continued on next page Table 44 – continued from previous page Sub-criterionDefinitionIndicators 24.3. Digital Divide Does the policy address the digital divide and ensure that access to AI technologies and their benefits is not limited by factors such as internet access, digital literacy, or device availability? 1. No consideration of the digital divide. 2. Digital divide is mentioned but not prioritized. 3. Some initiatives to address the digital divide. 4. Clear guidelines and programs to bridge the digital divide. 5. Comprehensive framework for addressing the digital divide, including provisions for infrastructure development, digital literacy training, and affordable access to devices and internet. Table 44: Social Equity Sub-criteria and Indicators (25) Public Trust. Sub-criterionDefinitionIndicators 25.1. Public Awareness and Education Does the policy include initiatives to raise public awareness and understanding of AI technologies, their potential benefits and risks? 1. No provisions for public awareness or education. 2. Public awareness is mentioned but not prioritized. 3. Some initiatives to raise public awareness. 4. Clear guidelines and programs for public education on AI. 5. Comprehensive public awareness and education campaigns, including resources for different stakeholder groups. 25.2. Public Engagement Does the policy provide opportunities for public engagement and consultation in the development and implementation of AI policies? 1. No mechanisms for public engagement. 2. Limited opportunities for public input. 3. Some mechanisms for public consultation. 4. Well-defined mechanisms for public engagement. 5. Multiple, robust mechanisms for public engagement, ensuring diverse perspectives are considered in policy-making. 25.3. Trust-Building Measures Does the policy promote trust-building measures, such as transparency, explainability, and accountability, in the design and deployment of AI systems? 1. No explicit trust-building measures. 2. Trust is mentioned but not addressed in detail. 3. Some guidelines for promoting trust. 4. Clear requirements for transparency, explainability, and accountability. 5. Comprehensive framework for building and maintaining public trust in AI, including independent audits, certification schemes, and redress mechanisms. Table 45: Public Trust Sub-criteria and Indicators References Adams, R.; Adeleke, F.; Alayande, A.; Abdella, S. E.; Florido, A.; Grossman, N.; and Junck, L. 2026. Global Index on Responsible AI: 2026 Report. arXiv:2607.14782. Administration, I. T. 2023. South Korea Semiconduc- tors. https://w.trade.gov/market-intelligence/south- korea-semiconductors. [Accessed 18-05-2025]. Alaga, J.; Schuett, J.; and Anderljung, M. 2024. A Grading Rubric for AI Safety Frameworks. arXiv:2409.08751. Alibaba Cloud. 2026. Elastic Compute Service Regions and Availability Zones. Allgeyer, T. 2023.AI Revolution in German AutomotiveIndustry:BoostingEfficiencyand Innovation.https://frenus.com/ai-revolution-in- german-automotive-industry-boosting-efficiency-and- innovation/. [Accessed 17-05-2025]. Almeida, P. G. R. d.; dos Santos Jr., C. D.; and Farias, J. S. 2020. Artificial Intelligence Regulation: A Meta- Framework for Formulation and Governance. In Pro- ceedings of the 53rd Hawaii International Conference on System Sciences. Amazon Web Services. 2026. Announcing Amazon EC2 Trn3 UltraServers for Faster, Lower-Cost Gener- ative AI Training. AWS What’s New. AP. 2023.Pentagon’s AI initiatives accel- erateharddecisionsonlethalautonomous weapons.https://economictimes.indiatimes. com/tech/technology/pentagons-ai-initiatives- accelerate-hard-decisions-on-lethal-autonomous- weapons/articleshow/105501657.cms.[Accessed 17-05-2025]. APAC Media. 2023.Amazon Web Services, Mi- crosoft Azure, Google Cloud Platform — the IndianCloudTrinity—apacnewsnetwork.com. https://apacnewsnetwork.com/2023/11/amazon-web- services-microsoft-azure-google-cloud-platform-the- indian-cloud-trinity/. [Accessed 18-05-2025]. Arnab Kumar, A. S., Punit Shukla; and Mahindru, T. 2024.National Strategy for Artificial Intelli- gence. https://w.niti.gov.in/sites/default/files/2023- 03/National-Strategy-for-Artificial-Intelligence.pdf. [Accessed 18-05-2025]. Atos. 2020. Atos powers Europe’s fastest supercom- puter at J ̈ ulich in Germany, the most energy-efficient system in the TOP100 — atos.net. https://atos.net/en/ 2020/press-release 20201117/atos-powers-europes- fastest-supercomputer-at-julich-in-germany-the-most- energy-efficient-system-in-the-top100.[Accessed 18-05-2025]. AWS.2022.AWSannouncesnewedge locationsinVietnam—aws.amazon.com. https://aws.amazon.com/about-aws/whats- new/2022/08/aws-new-edge-locations-vietnam/. [Accessed 16-05-2025]. AZB and Partners. 2024. MeitY Revises AI Advisory and Does Away with Government Approval Require- ment. Baig, A.; Khan, S.; and Gardezi, S. E. 2024.Sin- gapore’s Model AI Governance Framework - Se- curiti — securiti.ai.https://securiti.ai/singapore- model-ai-governance-framework/#: ∼ :text=Through% 20this%20approach%2C%20Singapore%20seeks,to% 20explore%20these%20aspects%20further. [Accessed 17-05-2025]. Baskoro, F. M. 2024.Sahabat-AI: Indosat and GoToLaunchIndonesia’sFirstOpen-Source AIEcosystemwithNvidia—jakartaglobe.id. https://jakartaglobe.id/tech/sahabatai-indosat- and-goto-launch-indonesias-first-opensource-ai- ecosystem-with-nvidia. [Accessed 16-05-2025]. Benn, D. 2025. Report highlights contributions of NHS AI Lab.https://w.nationalhealthexecutive.com/ articles/report-highlights-contributions-nhs-ai-lab. [Accessed 17-05-2025]. Bureau of Industry and Security, U.S. Department of Commerce. 2025.Department of Commerce An- nounces Rescission of Biden-Era Artificial Intelligence Diffusion Rule. BIS Press Release. Rescission an- nounced May 13, 2025. Butler, G. 2025.KT Cloud to establish AI Data Center demonstration center in Seoul, South Korea. https://w.datacenterdynamics.com/en/news/kt- cloud-to-establish-ai-data-center-demonstration- center-in-seoul-south-korea/. [Accessed 16-05-2025]. Cabinet Office, Government of Japan. 2019. Society 5.0: A Human-Centric Society. https://w8.cao.go.jp/ cstp/english/society5 0/index.html. [Accessed 15-05- 2025]. Cabinet Office, Government of Japan. 2025. Act on Promotion of Research and Development and Utiliza- tion of Artificial Intelligence-Related Technologies. Act No. 53 of 2025. Cai, F.; Zhang, J.; and Zhang, L. 2024. The Impact of Artificial Intelligence Replacing Humans in Making Human Resource Management Decisions on Fairness: A Case of Resume Screening. https://w.mdpi.com/ 2071-1050/16/9/3840. [Accessed 15-05-2025]. California State Government. 2023. California Con- sumer Privacy Act (CCPA) Update 2023. https://oag. ca.gov/privacy/ccpa. [Accessed 15-05-2025]. Center for AI Standards and Innovation; and National Institute of Standards and Technology. 2026. CAISI Evaluation of DeepSeek V4 Pro. NIST. Center for Strategic and International Studies. 2026. What to Know About Chinese AI Models. CSIS. Cerebras Systems. 2026. WSE-3: The Largest AI Chip Ever Built. Cerebras Product Documentation. Chang-won, L. 2021.Kakao’s AI research unit unveils hyperscale Korean-language AI platform based on GPT-3.https://w.ajupress.com/view/ 20211116134948085. [Accessed 16-05-2025]. Chia, O. 2024.Supercomputing sector gets $270millionboosttonurturespecialists, buildnext-gencomputers—straitstimes.com. https://w.straitstimes.com/tech/supercomputing- sector-gets-270-million-boost-to-nurture-specialists- build-next-gen-computers. [Accessed 17-05-2025]. Chun, J.; Schroeder de Witt, C.; and Elkins, K. 2024. Comparative Global AI Regulation: Policy Perspectives from the EU, China, and the US. arXiv:2410.21279. Cihon, P. 2019. Standards for AI governance: interna- tional standards to enable global coordination in AI re- search & development. Future of Humanity Institute. University of Oxford, 40(3): 340–342. CKGSB. 2025.China’s AI Leap: DeepSeek &GovtDriveAIDevelopment–CKGSB Knowledge—english.ckgsb.edu.cn.https: //english.ckgsb.edu.cn/knowledge/article/china-ai- leap-deepseek-govt-deive-ai-development/. [Accessed 18-05-2025]. Clifford, M.; and Department for Science, Innovation and Technology. 2025. AI Opportunities Action Plan. Technical report, UK Government. Published January 13, 2025. CNBC. 2025. Microsoft, Nvidia, Other Tech Giants Plan over $40 Billion of New AI Investments in UK. CNBC. Published September 16, 2025; US-UK Tech Prosperity Deal, GBP 31 billion in private AI/quantum investment. CNRS. 2025. Jean Zay supercomputer : France has increased its AI dedicated resources fourfold — cnrs.fr. https://w.cnrs.fr/en/press/jean-zay-supercomputer- france-has-increased-its-ai-dedicated-resources- fourfold. [Accessed 18-05-2025]. Cohen, J. 1968.Weighted Kappa: Nominal Scale Agreement with Provision for Scaled Disagreement or Partial Credit. Psychological Bulletin, 70(4): 213–220. Commission, E. 2021.Connecting Europe Facil- ity — commission.europa.eu.https://commission. europa.eu/funding-tenders/find-funding/eu-funding- programmes/connecting-europe-facility en. [Accessed 18-05-2025]. Council of the European Union. 2026. Artificial Intelli- gence: Council Gives Final Green Light to Simplify and Streamline Rules. Council of the EU Press Release. Fi- nal Council approval on June 29, 2026. Cyberspace Administration of China. 2022. Regula- tions on the Administration of Deep Synthesis Internet Information Services. Accessed: 2025-05-16. Cyberspace Administration of China. 2023. Interim Measures for the Management of Generative AI Ser- vices. Accessed: 2025-05-16. Cyberspace Administration of China and others. 2021. Provisions on the Administration of Algorithmic Rec- ommendations in Internet Information Services. Effec- tive March 1, 2022. Cyberspace Administration of China and others. 2025. Measures for the Labeling of AI-Generated and Syn- thetic Content. Joint measure of CAC, MIIT, the Min- istry of Public Security, and NRTA. Issued March 7, 2025 and effective September 1, 2025. Dat Quoc Nguyen, A. T. N. 2020.PhoBERT: Pre- trained language models for Vietnamese. https://arxiv. org/abs/2003.00744. [Accessed 17-05-2025]. de Fine Licht, K.; and de Fine Licht, J. 2020. Artificial intelligence, transparency, and public decision-making: Why explanations are key when trying to produce per- ceived legitimacy. AI & society, 35: 917–926. DeepSeek. 2025. DeepSeek-R1 Release. DeepSeek API Documentation. Released 20 January 2025. DeepSeek-AI. 2024. DeepSeek-V3 Technical Report. arXiv:2412.19437. Department for Science, Innovation and Technology. 2025. Tackling AI Security Risks to Unleash Growth and Deliver Plan for Change. GOV.UK. AI Safety Insti- tute renamed AI Security Institute, announced February 14, 2025. Department of Science and Technology. 2022. National Super Computing Mission — Department Of Science & Technology — dst.gov.in.https://dst.gov.in/ national-super-computing-mission. [Accessed 18-05- 2025]. Deradjat, A. A.; Timur, M.; Santoso, N. C.; and Kaur, D. P. 2025.Data Protection & Privacy 2025 - Indonesia — Global Practice Guides — Cham- bers and Partners — practiceguides.chambers.com. https://practiceguides.chambers.com/practice- guides/data-protection-privacy-2025/indonesia/trends- and-developments. [Accessed 17-05-2025]. Dharmaraj, S. 2025a.Bridging the AI Tal- entGap:Vietnam’sPathtoCompetitiveness. https://opengovasia.com/2025/02/06/bridging-the- ai-talent-gap-vietnams-path-to-competitiveness/. [Accessed 17-05-2025]. Dharmaraj,S.2025b.Thailand:National AICommitteeLaunchedtoDriveGrowth. https://opengovasia.com/2025/05/01/thailand-national- ai-committee-launched-to-drive-growth/.[Accessed 17-05-2025]. Digital India. 2026.IndiaAI Mission Expands AI Ecosystem with Affordable Compute and Startup Sup- port. Press Release, Ministry of Electronics and Infor- mation Technology. Digital Policy Alert. 2023.DPA — digitalpoli- cyalert.org. https://digitalpolicyalert.org/event/16382- adoption-of-the-circular-on-artificial-intelligence- ethics. [Accessed 17-05-2025]. drishtiias. 2020.Responsible AI for Youth.https: //w.drishtiias.com/daily-news-analysis/responsible- ai-for-youth. [Accessed 17-05-2025]. Dti. 2024. DTI launches National AI Strategy Roadmap 2.0 and Center for AI Research, positioning the Phil- ippines as a Center of Excellence in AI R&D. https: //w.dti.gov.ph/archives/news-archives/dti-launches- national-ai-strategy-roadmap-2-0-center-ai-research- positioning-philippines-center-excellence-ai-rampd/. [Accessed 17-05-2025]. DTI. 2024.DTI launches National AI Strategy Roadmap 2.0 and Center for AI Research, positioning the Philippines as a Center of Excellence in AI R&D — Department of Trade and Industry Philippines — dti.gov.ph.https://w.dti.gov.ph/archives/news- archives/dti-launches-national-ai-strategy-roadmap-2- 0-center-ai-research-positioning-philippines-center- excellence-ai-rampd/. [Accessed 16-05-2025]. dtiwebteam. 2024. DTI drives AI innovation with Na- tional Roadmap 2.0 and CAIR launch — Department of Trade and Industry Philippines — dti.gov.ph. https: //w.dti.gov.ph/archives/news-archives/dti-drives- ai-innovation-national-roadmap-2-0-cair-launch/. [Accessed 16-05-2025]. Dutta, S.; and Lanvin, B. 2022. The Network Readiness Index-Stepping into the new digital era How and why digital natives will change the world. https://download. networkreadinessindex.org/reports/nri 2022.pdf. [Ac- cessed 18-05-2025]. East Ventures. 2025. Artificial Intelligence in Indone- sia: The current state and its opportunities — east.vc. https://east.vc/news/insights/artificial-intelligence- in-indonesia-the-current-state-and-its-opportunities/. [Accessed 17-05-2025]. Engin, Z.; and Treleaven, P. 2019.Algorithmic Government: Automating Public Services and Sup- porting Civil Servants in using Data Science Tech- nologies. https://academic.oup.com/comjnl/article/62/ 3/448/5070384. [Accessed 15-05-2025]. European Commission. 2022.Digital Services Act (DSA).https://digital-strategy.ec.europa.eu/en/ policies/digital-services-act-package.[Accessed 15- 05-2025]. European Commission. 2024a. European AI Office. Eu- ropean Commission, Digital Strategy. Established by Commission Decision of 24 January 2024. European Commission. 2024b. Regulatory Framework Proposal on Artificial Intelligence.European Com- mission, Digital Strategy.AI Act (Regulation (EU) 2024/1689) entered into force 1 August 2024. European Commission. 2025a. AI Continent Action Plan. European Commission, Digital Strategy. Pub- lished 9 April 2025. European Commission. 2025b. EU Launches InvestAI Initiative to Mobilise EUR 200 Billion of Investment in Artificial Intelligence. European Commission Press Re- lease IP/25/467. Announced 11 February 2025 at Paris AI Action Summit. European Commission. 2025c.General-Purpose AI Code of Practice. European Commission, Digital Strat- egy. Finalized 10 July 2025; signatories include Ama- zon, Anthropic, Google, Microsoft, OpenAI, Mistral AI; Meta declined. European Commission. 2026. 2026 State of the Digital Decade Report. European Commission, Digital Strat- egy. European Parliamentary Research Service. 2022. The Artificial Intelligence Act: A risk-based approach to AI regulation.https://w.europarl.europa.eu/ RegData/etudes/BRIE/2022/733553/EPRS BRI(2022) 733553 EN.pdf. [Accessed 15-05-2025]. European Union. 2016. General Data Protection Reg- ulation (GDPR). https://eur-lex.europa.eu/eli/reg/2016/ 679/oj. [Accessed 15-05-2025]. European Union. 2021a.European High Per- formance Computing JU — European Union — european-union.europa.eu.https://european-union. europa.eu/institutions-law-budget/institutions-and- bodies/search-all-eu-institutions-and-bodies/european- high-performance-computing-joint-undertaking- eurohpc-ju en. [Accessed 18-05-2025]. European Union. 2021b.Proposal for a Reg- ulation laying down harmonised rules on arti- ficialintelligence(ArtificialIntelligenceAct). https://eur-lex.europa.eu/legal-content/EN/TXT/ ?uri=COM%3A2021%3A206%3AFIN.[Accessed 15-05-2025]. European Union. 2026. Regulation (EU) 2026/1744. Official Journal of the European Union. Eurostat.2025.Useofartificialintelli- genceinenterprises.https://ec.europa.eu/ eurostat/statistics-explained/index.php?title= Use ofartificialintelligenceinenterprises.[Ac- cessed 17-05-2025]. Eurostat. 2026.GDP and Main Components (Out- put, Expenditure and Income). Eurostat Database, ta- ble tec00001. EU27 2024 GDP at current prices: EUR 16.22 trillion. Fischer, S.-C.; Leung, J.; Anderljung, M.; O’keefe, C.; Torges, S.; Khan, S. M.; Garfinkel, B.; Dafoe, A.; Brundage, M.; Rey Ding, J.; et al. 2021. AI policy levers: A review of the US government’s tools to shape AI research, development, and deployment. Retrieved June, 1: 2022. FocusEconomics. 2024. Vietnam Archives — focus- economics.com.https://w.focus-economics.com/ countries/vietnam/. [Accessed 16-05-2025]. forDataEthics,C.;andInnovation.2021. CentreforDataEthicsandInnovationpub- lishesworldfirstroadmaptocatalysedevel- opment of AI assurance ecosystem — gov.uk. https://w.gov.uk/government/news/centre-for-data- ethics-and-innovation-publishes-world-first-roadmap- to-catalyse-development-of-ai-assurance-ecosystem. [Accessed 17-05-2025]. G, B. 2025.AWS opens Asia-Pacific head- quartersinSingapore—investmentmonitor.ai. https://w.investmentmonitor.ai/news/aws-opens- asia-pacific-headquarters-in-singapore/.[Accessed 16-05-2025]. Gaia-X European Association for Data and Cloud. 2025. Gaia-X Enters Season Two of Dataspaces and Digital Ecosystems with Summit 2025. Gaia-X official site. Gil, Y.; and Perrault, R. 2025. Artificial Intelligence In- dex Report 2025. https://hai-production.s3.amazonaws. com/files/hai aiindexreport2025.pdf. [Accessed 17- 05-2025]. Gomes, J. 2024. Singtel, NVIDIA to Bring Sovereign AI to Southeast Asia — blogs.nvidia.com. https://blogs. nvidia.com/blog/singtel-sovereign-ai/. [Accessed 17- 05-2025]. Google Cloud. 2026. Ironwood TPUs and New Axion- Based VMs for Your AI Workloads. Google Cloud Blog. Goovaerts, D. 2022.European cloud play- erslosingsharetoAWS,Microsoft,Google —fierce-network.com.https://w.fierce- network.com/cloud/european-cloud-players-losing- share-aws-microsoft-google. [Accessed 18-05-2025]. Government of India. 2023a. Digital Personal Data Protection Act, 2023. https://w.meity.gov.in/static/ uploads/2024/06/2bf1f0e9f04e6fb4f8fef35e82c42a5. pdf. [Accessed 15-05-2025]. Government of India. 2023b. Digital Personal Data Protection Act, 2023. https://w.meity.gov.in/static/ uploads/2024/06/2bf1f0e9f04e6fb4f8fef35e82c42a5. pdf. [Accessed 15-05-2025]. Government of Japan. 2025. Japan’s First AI Law: The Act on Promotion of Research and Development and Utilization of AI-Related Technologies. Highlighting Japan, Government Online. Hanh, V. N. 2024.Vietnam’s Digital Infrastructure Strategy Opens Investment Opportunities — vietnam- briefing.com. https://w.vietnam-briefing.com/news/ vietnams-digital-infrastructure-strategy-2025-new- opportunities-for-foreign-investors.html/.[Accessed 16-05-2025]. Harris, D. 2024. Japan Enhances AI Sovereignty With Advanced ABCI 3.0 Supercomputer.https://blogs. nvidia.com/blog/abci-aist/. [Accessed 17-05-2025]. Hattangudi,A.;Tharar,U.;andKukreja,A. 2024.morganstanley.com.https://w. morganstanley.com/im/publication/insights/articles/ article malaysianewfoundstabilitya4.pdf.[Accessed 16-05-2025]. Hendrycks, D.; Mazeika, M.; and Woodside, T. 2023.An Overview of Catastrophic AI Risks. arXiv:2306.12001. Hoa, N. Q. 2025.Vietnam’s AI Sector in 2025: Regulatory Frameworks & Investment Scope — vietnam-briefing.com.https://w.vietnam-briefing. com/news/vietnams-ai-sector-in-2025-regulatory- frameworks-and-opportunities-for-investors.html/. [Accessed 16-05-2025]. Holmstr ̈ om, J. 2022. From AI to Digital Transforma- tion: The AI Readiness Framework. Business Horizons, 65(3): 329–339. House of Commons Library. 2026. Gross Domestic Product (GDP): Economic Indicators. UK Parliament, House of Commons Library. UK nominal GDP 2025: approx. GBP 3.03 trillion. HPCwire. 2024.Supercomputer LUMI Remains as One of the Leading AI Platforms for Science and Innovation.https://w.hpcwire.com/off-the- wire/supercomputer-lumi-remains-as-one-of-the- leading-ai-platforms-for-science-and-innovation/. [Accessed 18-05-2025]. Huawei. 2025. Groundbreaking SuperPoD Intercon- nect: Leading a New Paradigm for AI Infrastructure. Huawei Newsroom, Keynote by Xu Zhijun. IEP. 2024.economicsandpeace.org.https: //w.economicsandpeace.org/wp-content/uploads/ 2024/06/GPI-2024-web.pdf. [Accessed 17-05-2025]. IMF. 2020.The Philippines: A Good Time to Expand the Infrastructure Push — imf.org. https://w.imf.org/en/News/Articles/2020/02/06/ na020620the-philippines-a-good-time-to-expand-the- infrastructure-push. [Accessed 16-05-2025]. IMF. 2025a. IMF Executive Board Concludes 2025 Article IV Consultation with Malaysia — imf.org. https://w.imf.org/en/News/Articles/2025/03/02/ pr25050-malaysia-imf-executive-board-concludes- 2025-article-iv-consultation. [Accessed 16-05-2025]. IMF. 2025b. imf.org. https://w.imf.org/external/ datamapper/profile/PHL. [Accessed 16-05-2025]. Ines Lin, D. A., Taipei; Charlene Chen. 2024. SEA cloud free-for-all intensifies as AWS expands new Malaysia Region — digitimes.com.https: //w.digitimes.com/news/a20240823PD210/aws- amazon-asia-microsoft-subsidiary.html.[Accessed 16-05-2025]. Infocomm Media Development Authority and AI Ver- ify Foundation. 2024. Model AI Governance Frame- work for Generative AI. IMDA/AI Verify Foundation. Finalized 30 May 2024. Infocomm Media Development Authority, Singapore. 2026.New Model AI Governance Framework for Agentic AI. IMDA Press Release. Launched 22 Jan- uary 2026 at WEF Davos. Insights,S.2023.JapanArtificialIntel- ligenceMarketInsightsForecaststo2032. https://w.sphericalinsights.com/reports/japan- artificial-intelligence-ai-market.[Accessed 17-05- 2025]. International Monetary Fund. 2025.IMF Executive Board Concludes 2025 Article IV Consultation with In- dia. IMF Press Release No. 25/392. International Monetary Fund. 2026. World Economic Outlook Database: South Korea Nominal GDP. IMF Datamapper. ITA. 2023. United Kingdom Artificial Intelligence Mar- ket 2023 — trade.gov. https://w.trade.gov/market- intelligence/united-kingdom-artificial-intelligence- market-2023. [Accessed 17-05-2025]. ITA. 2024.Malaysia - Digital Economy — trade.gov. https://w.trade.gov/country-commercial- guides/malaysia-digital-economy.[Accessed 16-05- 2025]. ITA. 2025.Indonesia Digital Transformation — trade.gov. https://w.trade.gov/market-intelligence/ indonesia-digital-transformation.[Accessed 16-05- 2025]. Jackson, A. 2025.How CoreWeave is Work- ing to Accelerate UK AI Infrastructure.https:// datacentremagazine.com/data-centres/how-coreweave- is-working-to-accelerate-ukai-infrastructure.[Ac- cessed 16-05-2025]. Japanese Ministry of Economy, Trade and Indus- try. 2020.Social Principles of Human-Centric AI. https://w.cas.go.jp/jp/seisaku/jinkouchinou/pdf/ humancentricai.pdf. [Accessed 15-05-2025]. Jevinger, A.; Zhao, C.; and Persson, J. e. a. 2024.Artificial intelligence for improving public transport: a mapping study - Public Transport — link.springer.com. https://link.springer.com/article/10. 1007/s12469-023-00334-7. [Accessed 15-05-2025]. Ji, J.; Qiu, T.; Chen, B.; Zhang, B.; Lou, H.; Wang, K.; Duan, Y.; He, Z.; Vierling, L.; Hong, D.; Zhou, J.; Zhang, Z.; Zeng, F.; Dai, J.; Pan, X.; Ng, K. Y.; O’Gara, A.; Xu, H.; Tse, B.; Fu, J.; McAleer, S.; Yang, Y.; Wang, Y.; Zhu, S.-C.; Guo, Y.; and Gao, W. 2025. AI Align- ment: A Comprehensive Survey. arXiv:2310.19852. Jon, W. J. 2026. South Korea’s AI Framework Act Fo- cuses on Rights and Safety. Nature, 651(8104): 274. Karen. 2022.Philippines’ Cloud Market Continues to Expand - Telecom Review Asia — telecomre- viewasia.com.https://w.telecomreviewasia.com/ news/featured-articles/3102-philippines-cloud-market- continues-to-expand/. [Accessed 16-05-2025]. Keat, D. H. S. 2024.DPM Heng Swee Keat at the Launch of the National Supercomputing Centre Singapore’s Aspire 2A and Aspire 2A+ Super- computer — pmo.gov.sg.https://w.pmo.gov.sg/ Newsroom/DPM-Heng-Swee-Keat-at-the-Launch- of-The-National-Supercomputing-Centre-Singapore. [Accessed 16-05-2025]. Korolov, S. 2025. Where in London Lies the Biggest Potential for AI Innovations?https://railsware.com/ blog/ai-london-evolution/. [Accessed 16-05-2025]. Krause, D.; and Th ̈ ornig, P. 2016. JURECA: General- purpose supercomputer at J ̈ ulich Supercomputing Cen- tre. Krippendorff, K. 2018. Content Analysis: An Introduc- tion to Its Methodology. SAGE Publications, 4 edition. ISBN 9781506395678. Kumar, A. 2024.TCS’ AI pipeline at $1.5bn; fo- cus on talent development, AI engagement.https: //w.business-standard.com/companies/news/tcs- to-invest-1-5-bn-in-ai-to-transform-hiring-process- work-environment-124071700562 1.html.[Accessed 16-05-2025]. Kurilla, M. 2025. What Is the CHIPS Act? https:// w.cfr.org/in-brief/what-chips-act. [Accessed 17-05- 2025]. Leesa-nguansuk, S. 2025. Siam.AI to invest B70bn to expand Thailand AI infrastructure.https://w. bangkokpost.com/business/general/2960476/siam-ai- invest-70-billion-to-expand-thailand-ai-infrastructure. [Accessed 17-05-2025]. Lehdonvirta, V. 2024. Compute North vs. Compute South: The Uneven Possibilities of Compute-based AI Governance Around the Globe — Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society — ojs.aaai.org.https://ojs.aaai.org/index.php/AIES/ article/view/31683. [Accessed 17-05-2025]. Li, Y. 2022. Research and application of deep learning in image recognition. In 2022 IEEE 2nd international conference on power, electronics and computer appli- cations (ICPECA), 994–999. IEEE. Lim, D. 2024. Determinants of Socially Responsible AI Governance. Duke L. & Tech. Rev., 25: 183. Loyola, J. A. 2024.Converge, Supermicro to build green AI data centers — mb.com.ph. https://mb.com.ph/2024/7/4/converge-supermicro- to-build-green-ai-data-centers.[Accessed 16-05- 2025]. MacroPolo. 2025.The Global AI Talent Tracker 3.0.MacroPolo Archive, Paulson Institute.Based on NeurIPS 2024, ICML 2024, ICLR 2025 data; finds only 11 percent retention of China-educated elite AI re- searchers, down from 16 percent in 2019. Malaysiakini. 2024.Launching of the MyGenom Project: Advancing Precision Medicine and Genomic Research in Malaysia. https://w.malaysiakini.com/ announcement/717803. [Accessed 17-05-2025]. Maslej, N.; Fattorini, L.; Perrault, R.; Gil, Y.; Parli, V.; Kariuki, N.; Capstick, E.; Reuel, A.; Brynjolfsson, E.; Etchemendy, J.; Ligett, K.; Lyons, T.; Manyika, J.; Niebles, J. C.; Shoham, Y.; Wald, R.; and Walsh, T. 2025a.The AI Index 2025 Annual Report.Tech- nical report, AI Index Steering Committee, Institute for Human-Centered AI, Stanford University, Stanford, CA. Also available as arXiv:2504.07139; released 7 April 2025. Maslej, N.; Fattorini, L.; Perrault, R.; Gil, Y.; Parli, V.; Kariuki, N.; Capstick, E.; Reuel, A.; Brynjolfsson, E.; Etchemendy, J.; Ligett, K.; Lyons, T.; Manyika, J.; Niebles, J. C.; Shoham, Y.; Wald, R.; Walsh, T.; Ham- rah, A.; Santarlasci, L.; Lotufo, J. B.; Rome, A.; Shi, A.; and Oak, S. 2025b. The AI Index 2025 Annual Report. Accessed: 2025-05-15. McFaul, C.; Chahal, H.; Gelles, R.; and Konaev, M. 2023.Assessing South Korea’s AI Ecosys- tem. https://cset.georgetown.edu/publication/assessing- south-koreas-ai-ecosystem/. [Accessed 17-05-2025]. Mehra, S. 2024.Indonesia Cloud Computing Mar- ket Outlook to 2028 — kenresearch.com.https: //w.kenresearch.com/industry-reports/indonesia- cloud-computing-market. [Accessed 16-05-2025]. Merritt, R. 2021. Asia’s Rising Star: VinAI Advances Innovation with Vietnam’s Most Powerful AI Super- computer — blogs.nvidia.com.https://blogs.nvidia. com/blog/ai-superpod-vinai/. [Accessed 16-05-2025]. Meta. 2023.Europe - Meta Data Centers — dat- acenters.atmeta.com.https://datacenters.atmeta.com/ europe/. [Accessed 18-05-2025]. MIDA. 2024a.Driving Malaysia’s solar power adoption through commercial, residential properties — mida.gov.my.https://w.mida.gov.my/mida- news/driving-malaysias-solar-power-adoption- through-commercial-residential-properties/.[Ac- cessed 16-05-2025]. MIDA. 2024b. Nvidia-YTL tie-up to boost Malaysia’s AI infrastructure - Tengku Zafrul — mida.gov.my. https://w.mida.gov.my/mida-news/nvidia-ytl- tie-up-to-boost-malaysias-ai-infrastructure-tengku- zafrul/. [Accessed 16-05-2025]. Mikhaylov, S. J.; Esteve, M.; and Campion, A. 2018. Artificial intelligence for the public sector: opportuni- ties and challenges of cross-sector collaboration. Philo- sophical transactions of the royal society a: mathe- matical, physical and engineering sciences, 376(2128): 20170357. Miller, K.; and Gelles, R. 2024. The NAIRR Pilot: Es- timating Compute. https://cset.georgetown.edu/article/ the-nairr-pilot-estimating-compute/. [Accessed 16-05- 2025]. Min-gwan, S.; and Su-hyeon, P. 2025. South Korea bets $1.3B on AI hardware, critics say talent overlooked. https://w.chosun.com/english/industry-en/2025/ 04/22/6R32I3L5RNHMJIQLA7U3MYE3DE/#: ∼ : text=South%20Korea’s%20government%20unveiled% 20a,artificial%20intelligence%20(AI)%20capabilities. [Accessed 17-05-2025]. Ministry of Economy, T.; and Industry. 2022. Gov- ernance Guidelines for Implementation of AI Prin- ciples.https://w.meti.go.jp/shingikai/mono info service/aishakaijisso/pdf/202201282.pdf. [Accessed 17-05-2025]. Ministry of Economy, Trade and Industry and Ministry of Internal Affairs and Communications, Japan. 2024. AI Guidelines for Business Version 1.0. Ministry of Education India. 2022. Artificial Intelli- gence Research at IITs. https://asean-iit.in/artificial- intelligence-research-at-iits/. [Accessed 17-05-2025]. Ministry of Electronics and Information Technology, Government of India. 2025.India AI Governance Guidelines. Government report. Released November 5, 2025. Ministry of Health Singapore. 2007. Human Biomed- ical Research Act (HBRA). https://sso.agc.gov.sg/Act/ HBRA2015. [Accessed 15-05-2025]. Ministry of Health Singapore. 2022.Health- careCybersecurityEssentialsGuidelines. https://w.healthinfo.gov.sg/files/MOH CirNo 85202304Dec2023CyberandDataSecurity GuidelinesforHealthcareProvidersAnnexA.pdf. [Accessed 15-05-2025]. Ministry of Health Singapore; Health Sciences Au- thority; and Integrated Health Information Systems. 2023. Artificial Intelligence in Healthcare Guidelines. https://isomer-user-content.by.gov.sg/3/9c0db09d- 104c-48af-87c9-17e01695c67c/1-0-artificial-in- healthcare-guidelines-(aihgle) publishedoct21.pdf. [Accessed 15-05-2025]. Ministry of Industry. 2018.Indonesia’s Fourth In- dustrial Revolution.https://sea-vet.net/images/seb/ initiatives/appendixfile/570/making-indonsia-40- bppi.pdf. [Accessed 17-05-2025]. Ministry of Science and ICT, Republic of Korea. 2020. National Guidelines for AI Ethics. Ministry of Trade and Industry, Singapore. 2026. Singa- pore’s GDP Grew by 5.7 Per Cent in the Fourth Quarter of 2025 and by 4.8 Per Cent in 2025. MTI Press Re- lease. Mohamed, Y. A.; Khanan, A.; Bashir, M.; Mohamed, A. H. H.; Adiel, M. A.; and Elsadig, M. A. 2024. The impact of artificial intelligence on language translation: a review. Ieee Access, 12: 25553–25579. Monetary Authority of Singapore. 2021.FEAT Principles for Responsible AI.https://w.mas. gov.sg/ ∼ /media/MAS/News%20and%20Publications/ Monographs%20and%20Information%20Papers/ FEAT%20Principles%20Final.pdf. [Accessed 15-05- 2025]. Mothership. 2025. Budget 2025: S$150 Million for En- terprise Compute Initiative to Help Firms Adopt AI. Mothership.sg. Narioka, K. 2024.SoftBank Corp. to Build Large-Scale AI Data Center in Osaka.https: //w.wsj.com/business/telecom/softbank-corp- to-build-large-scale-ai-data-center-in-osaka- 7000a213?utm source=chatgpt.com.[Accessed 17-05-2025]. National Bureau of Statistics of China. 2025. Final Ver- ified Accounting Results of Gross Domestic Product for 2024. National Bureau of Statistics of China. Final re- vised 2024 GDP: 134.8066 trillion yuan (approx. USD 19.16 trillion), 5.0 percent growth. National Institute of Standards and Technology (NIST). 2023. Artificial Intelligence Risk Management Frame- work (AI RMF) 1.0. https://w.nist.gov/itl/ai-risk- management-framework. [Accessed 15-05-2025]. National People’s Congress Standing Committee. 2025. Amendment to the Cybersecurity Law of the People’s Republic of China.NPC Official.Approved late October 2025, effective 1 January 2026; includes AI safety/development article. NeGD. 2020.YUVAi - Youth for Unnati & Vikas with AI — responsibleaiforyouth.negd.in. https: //responsibleaiforyouth.negd.in/home.[Accessed 17- 05-2025]. Nhandan. 2025. Vietnam National University, Hanoi and Tsinghua University collaborate on advanced AI research.https://en.nhandan.vn/vietnam-national- university-hanoi-and-tsinghua-university-collaborate- on-advanced-ai-research-post144699.html. [Accessed 17-05-2025]. NITI Aayog. 2018.National Strategy for Arti- ficial Intelligence.https://w.niti.gov.in/sites/ default/files/2023-03/National-Strategy-for-Artificial- Intelligence.pdf. [Accessed 15-05-2025]. NITIAayog,GovernmentofIndia.2018. NationalStrategyforArtificialIntelligence. https://w.niti.gov.in/sites/default/files/2023- 03/National-Strategy-for-Artificial-Intelligence.pdf. [Accessed 15-05-2025]. NSTDA. 2019.Research Strategy - NSTDA Eng — nstda.or.th.https://w.nstda.or.th/en/research/ research-strategy.html. [Accessed 16-05-2025]. NVIDIA. 2022.NVIDIA Signs Agreement to Develop AI Talent, Support Industry Research in Indonesia — nvidia.com. https://w.nvidia.com/en- sg/news/nvidia-signs-agreement-to-develop-ai-talent- support-industry-research-in-indonesia/.[Accessed 16-05-2025]. NVIDIA. 2024.NVIDIA to Open Vietnam R&D CentertoBolsterAIDevelopment—nvidi- anews.nvidia.com.https://nvidianews.nvidia.com/ news/nvidia-to-open-vietnam-r-d-center-to-bolster-ai- development. [Accessed 16-05-2025]. NVIDIA. 2025a. ABCI 3.0: Japan’s National AI Cloud Infrastructure at AIST. NVIDIA Blog. NVIDIA. 2025b. South Korea, Government and In- dustrial Giants Build AI Infrastructure and Ecosystem. NVIDIA Newsroom. Announced at APEC Summit, Gyeongju, 31 October 2025. NVIDIA. 2026. NVIDIA Kicks Off the Next Genera- tion of AI with Rubin, Six New Chips, One AI Super- computer. NVIDIA Newsroom. Obeng, S.; Iyelolu, T. V.; Akinsulire, A. A.; and Ide- mudia, C. 2024. Utilizing machine learning algorithms to prevent financial fraud and ensure transaction secu- rity. World Journal of Advanced Research and Reviews, 23(1): 1972–1980. OECD. 2024. G7 toolkit for Artificial Intelligence in the public sector — OECD. Olukoya, O. 2022. Assessing Frameworks for Elicit- ing Privacy and Security Requirements from Laws and Regulations. Computers & Security, 117: 102697. Oschinski, M.; Feldgoise, J.; Neufeld, J.; and Alcorn, S. 2025. Strengthening America’s AI Workforce. https: //ifp.org/strengthening-americas-ai-workforce/.[Ac- cessed 17-05-2025]. Oxford Insights. 2025. Government AI Readiness Index 2025. Technical report, Oxford Insights. Park, E.-M. 2024.SK Hynix looks to ex- pand AI memory leadership with new NAND chip.https://w.kedglobal.com/korean- chipmakers/newsView/ked202405090014.[Accessed 16-05-2025]. Parliament, U. 2018.Data Protection Act 2018. https://w.legislation.gov.uk/ukpga/2018/12/ contents/enacted. [Accessed 17-05-2025]. Perry, B.; and Uuk, R. 2019. AI Governance and the Policymaking Process: Key Considerations for Reduc- ing AI Risk. Big Data and Cognitive Computing, 3(2): 26. PersonalDataProtectionCommissionSinga- pore. 2020.Model AI Governance Framework. https://w.pdpc.gov.sg/Help-and-Resources/2020/ 01/Model-AI-Governance-Framework.[Accessed 15-05-2025]. Personal Information Protection Commission, Republic of Korea. 2023. Personal Information Protection Act, Article 37-2 and 2023 Amendments. Peterson, D. L. 2024.India’s AI ambitions: Can public-private partnerships lead the way?https: //w.spglobal.com/en/research-insights/special- reports/india-forward/indias-ai-ambitions-can-public- private-partnerships-lead-the-way. [Accessed 18-05- 2025]. Pi, Y. 2021. Machine learning in governments: Benefits, challenges and future directions. JeDEM-eJournal of eDemocracy and Open Government, 13(1): 203–219. Press Information Bureau, Government of India. 2025. Digital Personal Data Protection Rules, 2025 Notified. PIB Press Note. Notified 13–14 November 2025. Rahman, S.; Islam, M.; Hossain, I.; and Ahmed, A. 2024. Utilizing AI and data analytics for optimizing re- source allocation in smart cities: A US based study. In- ternational journal of artificial intelligence, 4(07): 70– 95. Reddy, S.; Rogers, W.; Makinen, V.-P.; Coiera, E.; Brown, P.; Wenzel, M.; Weicken, E.; Ansari, S.; Mathur, P.; Casey, A.; and Kelly, B. 2021. Evaluation Frame- work to Guide Implementation of AI Systems into Healthcare Settings. BMJ Health and Care Informat- ics, 28(1): e100444. Republic of Korea. 2025. Framework Act on the Devel- opment of Artificial Intelligence and the Creation of a Foundation for Trust. Act No. 20676. Effective January 22, 2026. Research, U.; and Innovation. 2021. ARCHER2 High performance computing for national research. https: //w.discover.ukri.org/archer2/index.html. [Accessed 16-05-2025]. RFA. 2025. China, US, Nvidia Chips: Export Restric- tions and Reversals. Radio Free Asia. Rough, E.; and House of Commons Library. 2026. AI Regulation in the UK (Research Briefing CBP-10003). Technical report, UK Parliament, House of Commons Library. Published June 10, 2026; covers DSIT’s Octo- ber 21, 2025 Blueprint for AI Regulation. Sakdhnagool, P.; Kunaseth, M.; Piyatumrong, A.; Chompoonut Rungnim, V. J.; Udomsiripinij, W.; Tharatipayakul, K.; and Takdanai Suwan, P. S. 2021. TARA: A Year in Review: Putt Sakdhnagool, Man- aschai Kunaseth, Apivadee Piyatumrong, Chompoonut Rungnim, Viwan Jarerattanachat, Wirote Udom- siripinij, Kittithorn Tharatipayakul, Takdanai Suwan, Piyawut Srichaikul — Thai Journal of Mathematics — thaijmath2.in.cmu.ac.th. https://thaijmath2.in.cmu.ac. th/index.php/thaijmath/article/view/1192.[Accessed 16-05-2025]. Samuel, P. 2021.Vietnam’s Digital Transforma- tion Plan Through 2025 — vietnam-briefing.com. https://w.vietnam-briefing.com/news/vietnams- digital-transformation-plan-through-2025.html/. [Accessed 16-05-2025]. Shainer, G. 2021.NVIDIA Accelerates World’s First TOP500 Academic Cloud-Native Supercom- puter to Advance Research at Cambridge Univer- sity. https://blogs.nvidia.com/blog/csd3-cloud-native- supercomputer-cambridge-university/. [Accessed 16- 05-2025]. Sharma,P.2024.UKUniversitiesLead- inginDataScienceandAIPrograms. https://w.analyticsinsight.net/data-science/uk- universities-leading-in-data-science-and-ai-programs. [Accessed 17-05-2025]. Singh, T.; Shivraj, A.; Madaan, S.; and Nigam, M. 2025.Unlocking AI’s Potential in In- diaTransformingAgricultureandHealthcare. https://w.bcg.com/publications/2025/india- unlocking-ai-potential-in-india-transforming- agriculture-and-healthcare. [Accessed 17-05-2025]. Singhania, S. 2024. The Role of AI in Shaping India’s Digital Economy. https://indiaai.gov.in/article/the-role- of-ai-in-shaping-india-s-digital-economy.[Accessed 17-05-2025]. Slattery, P.; Saeri, A. K.; Grundy, E. A. C.; Graham, J.; Noetel, M.; Uuk, R.; Dao, J.; Pour, S.; Casper, S.; and Thompson, N. 2025. The AI Risk Repository: A Com- prehensive Meta-Review, Database, and Taxonomy of Risks From Artificial Intelligence. arXiv:2408.12622. Smart Nation Singapore. 2019.National AI Strat- egy.https://w.smartnation.gov.sg/nais/#: ∼ : text=Singapore%20National%20AI%20Strategy%202. 0%20(NAIS%202.0),-To%20propel%20ourselves& text=The%20expanded%20landscape%20of% 20opportunities,an%20accessory%20or%20a% 20luxury. [Accessed 17-05-2025]. Standing Committee of the National People’s Congress of China. 2016. Cybersecurity Law of the People’s Re- public of China. Standing Committee of the National People’s Congress of China. 2021a. Data Security Law of the People’s Republic of China. Standing Committee of the National People’s Congress of China. 2021b. Personal Information Protection Law of the People’s Republic of China. Stanford Institute for Human-Centered Artificial Intel- ligence. 2025.The 2025 AI Index Report, Chapter 1: Research and Development. Stanford HAI. China 23.2 percent of AI publications, 22.6 percent of cita- tions (2023 data). Stanford Institute for Human-Centered Artificial Intel- ligence. 2026. The 2026 AI Index Report. Technical report, Stanford University. Published 13 April 2026. Stanly, M. 2024. India’s AI Talent Pool to Grow to 1.25 Million by 2027: Nasscom-Deloitte India Report. https://indiaai.gov.in/article/india-s-ai-talent-pool- to-grow-to-1-25-million-by-2027-nasscom-deloitte- india-report. [Accessed 17-05-2025]. State Council of the People’s Republic of China. 2024a. China Invests Over 6.1 Billion USD in Major Comput- ing Hubs. State Council of the People’s Republic of China. 2024b. Regulations on Network Data Security Management. Effective January 1, 2025. State Council of the People’s Republic of China. 2025. Opinions on Deeply Implementing the ”Artificial Intel- ligence Plus” Action (Guo Fa [2025] No. 11). gov.cn official English site. Approved 31 July 2025, published 26-27 August 2025. StateScoop. 2025. State AI Law Moratorium Omitted from 2026 Defense Bill, but Trump Prepares Executive Order. StateScoop. See also Executive Order 14365, “Ensuring a National Policy Framework for Artificial Intelligence,” signed December 11, 2025. Staufer, L.; et al. 2026. The 2025 AI Agent Index: Doc- umenting Technical and Safety Features of Deployed Agentic AI Systems. arXiv preprint arXiv:2602.17753. Accepted to ACM FAccT 2026. Stelling, L.; Murray, M.; Galizzi, B.; Schaffelder, M.; Campos, S.; and Papadatos, H. 2025.Eval- uating AI Providers’ Frontier Safety Frameworks. arXiv:2512.01166. Stix, C. 2021. Actionable Principles for Artificial In- telligence Policy: Three Pathways — doi.org. https: //doi.org/10.48550/arXiv.2102.12406.[Accessed 15- 05-2025]. Tachev, V. 2024. Renewable Energy in the Philippines – Current State and Future Roadmap — energy- tracker.asia.https://energytracker.asia/renewable- energy-in-the-philippines-current-state-and-future- roadmap/. [Accessed 16-05-2025]. Takano, R.; Takizawa, S.; Tanimura, Y.; and Nakada, H. 2024. ABCI 3.0: Evolution of the leading AI infras- tructure in Japan. https://arxiv.org/html/2411.09134v1. [Accessed 16-05-2025]. Tambiama Madiega with Rafał Ilnicki; Graph- ics: Eulalia Claros, M. R. S. 2024.AI in- vestment: EU and global indicators.https: //w.europarl.europa.eu/RegData/etudes/ATAG/ 2024/760392/EPRS ATA(2024)760392EN.pdf. [Ac- cessed 17-05-2025]. Tazrout, Z. 2021. Vietnam: Launch of an international research center on artificial intelligence — actuia.com. https://w.actuia.com/en/news/vietnam-launch- of-an-international-research-center-on-artificial- intelligence/. [Accessed 16-05-2025]. TechCrunch; and Wiggers, K. 2025. As US and UK Refuse to Sign the AI Action Summit Statement, Other Countries Commit to AI Governance.TechCrunch. Paris AI Action Summit, 10–11 February 2025. Tencent Cloud. 2026. Regions and Zones. The Edge Singapore. 2024. Singapore to Invest Over $1 Billion in AI Over Next Five Years: Budget 2024. The Edge Singapore. Budget delivered 16 February 2024. The White House. 2025a. Initial Rescissions of Harm- ful Executive Orders and Actions. Presidential Actions, The White House. Executive Order 14148, signed Jan- uary 20, 2025; published as 90 FR 8237. The White House. 2025b. Removing Barriers to Amer- ican Leadership in Artificial Intelligence. Presidential Actions, The White House. Executive Order 14179, signed January 23, 2025. The White House / Office of Science and Technology Policy. 2025. Winning the Race: America’s AI Action Plan. Technical report, The White House. Released July 23, 2025. TOP500.org. 2026a.LineShine Debuts at No. 1 as the TOP500 Enters a New Global Exascale Era. TOP500.org, 67th Edition List.June 2026 list; LineShine (Shenzhen) achieves 2.198 exaFLOPS HPL, first Chinese No. 1 since 2017. TOP500.org. 2026b.TOP500 List, June 2026. TOP500.org. Truby, J.; Brown, R. D.; Ibrahim, I. A.; and Parellada, O. C. 2022. A sandbox approach to regulating high-risk artificial intelligence applications. European Journal of Risk Regulation, 13(2): 270–294. UK Department for Science, Innovation and Technol- ogy. 2023a. The Bletchley Declaration by Countries Attending the AI Safety Summit, 1–2 November 2023. GOV.UK. Signed by 28 countries and the EU on 1 November 2023. UKDepartmentforScience,Innovationand Technology.2023b.APro-InnovationAp- proach to AI Regulation: Government Response. https://w.gov.uk/government/publications/ai- regulation-a-pro-innovation-approach/white-paper. [Accessed 15-05-2025]. UK Department for Science, Innovation and Technol- ogy; and Government of the Republic of Korea. 2024. Seoul Declaration for Safe, Innovative and Inclusive AI: AI Seoul Summit, 21 May 2024. GOV.UK. Compan- ion: Frontier AI Safety Commitments, signed by 16 AI companies. UK Government. 2021.National AI Strategy. https://w.gov.uk/government/publications/national- ai-strategy. [Accessed 15-05-2025]. UK Government Digital Service. 2025. Algorithmic Transparency Recording Standard: Guidance for Public Sector Bodies. UK Intellectual Property Office. 2022. Artificial Intelli- gence and Intellectual Property: Copyright and Patents, Government Response to Consultation. UKRI. 2022. Innovate UK strategic delivery plan 2022 to 2025 — ukri.org. https://w.ukri.org/publications/ innovate-uk-strategic-delivery-plan/innovate-uk- strategic-delivery-plan-2022-to-2025/.[Accessed 17-05-2025]. under the Ministry of Digital Economy, N. A. I. C. N.; and (MDES), S. 2021a.Thailand’s Na- tional AI Strategy and Action Plan (2022-2027) —DigitalWatchObservatory—dig.watch. https://dig.watch/resource/thailands-national-ai- strategy-and-action-plan-2022-2027.[Accessed 16-05-2025]. under the Ministry of Digital Economy, N. A. I. C. N.; and (MDES), S. 2021b.Thailand’s Na- tional AI Strategy and Action Plan (2022-2027) —DigitalWatchObservatory—dig.watch. https://dig.watch/resource/thailands-national-ai- strategy-and-action-plan-2022-2027.[Accessed 16-05-2025]. under the Ministry of Digital Economy, N. A. I. C. N.; and (MDES), S. 2021c.Thailand’s Na- tional AI Strategy and Action Plan (2022-2027) —DigitalWatchObservatory—dig.watch. https://dig.watch/resource/thailands-national-ai- strategy-and-action-plan-2022-2027.[Accessed 16-05-2025]. UNESCO. 2023. Readiness Assessment Methodology: A Tool of the Recommendation on the Ethics of Artifi- cial Intelligence. University of Bristol. 2025. 2025: Isambard-AI is 11th Fastest Supercomputer in the World. University of Bris- tol Research News. U.S. Bureau of Economic Analysis. 2026. GDP (Third Estimate), Industries, Corporate Profits, State GDP, and State Personal Income, 4th Quarter and Year 2025. U.S. Bureau of Economic Analysis. Full-year 2025 current- dollar GDP approx. $30.5–30.8 trillion. U.S. Congress. 2020. National Artificial Intelligence Initiative Act of 2020.https://w.congress.gov/ bill/116th-congress/house-bill/6216. [Accessed 15-05- 2025]. US Department of State. 2024.Technical Difficul- ties — state.gov. https://w.state.gov/reports/2024- investment-climate-statements/vietnam/.[Accessed 16-05-2025]. van Noordt, C.; and Misuraca, G. 2022. Artificial in- telligence for the public sector: results of landscap- ing the use of AI in government across the European Union. https://w.sciencedirect.com/science/article/ abs/pii/S0740624X22000478?dgcid=author. [Accessed 15-05-2025]. Vietnamplus.2025.Malaysialaunchesna- tionaldigitalpedagogytrainingprogramme. https://en.vietnamplus.vn/malaysia-launches- national-digital-pedagogy-training-programme- post317837.vnp. [Accessed 17-05-2025]. Viettel. 2022. viettel.com.vn. https://viettel.com.vn/en/ news-events/news/viettel-enters-ai-collaboration-with- nvidia/. [Accessed 16-05-2025]. VinAI. 2024.VinAI Recognized as “Beacon of Innovation” in Deloitte’s Edges of Southeast Asia 2024 - VinAI — vinai.io. https://w.vinai.io/vinai- recognized-as-beacon-of-innovation-in-deloittes- edges-of-southeast-asia-2024/.[Accessed 16-05- 2025]. Webster, G.; Creemers, R.; Kania, E.; and Tri- olo, P. 2017.Full Translation: China’s ’New GenerationArtificialIntelligenceDevelopment Plan’ (2017) - DigiChina — digichina.stanford.edu. https://digichina.stanford.edu/work/full-translation- chinas-new-generation-artificial-intelligence- development-plan-2017/. [Accessed 17-05-2025]. Weiyue Wu, S. L. 2023. A Comprehensive Review and Systematic Analysis of Artificial Intelligence Regula- tion Policies. https://arxiv.org/pdf/2307.12218. [Ac- cessed 15-05-2025]. White House. 2022.Blueprint for an AI Bill of Rights.https://bidenwhitehouse.archives.gov/ostp/ai- bill-of-rights/. [Accessed 15-05-2025]. White House. 2023.Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence.https://w.federalregister. gov/documents/2023/11/01/2023-24283/safe-secure- and-trustworthy-development-and-use-of-artificial- intelligence. [Accessed 15-05-2025]. Williams, J. 2023.Deutsche Telekom joins project to build future of European Cloud-Edge infrastruc- ture.https://w.telekom.com/en/media/media- information/archive/edge-cloud-infrastructure-project- 1054872. [Accessed 16-05-2025]. Wire, B. 2024.BDx Data Centers Introduces In- donesia’s First Sovereign AI Data Center, Built With NVIDIA Accelerated Computing — business- wire.com. https://w.businesswire.com/news/home/ 20241129476471/en/BDx-Data-Centers-Introduces- Indonesias-First-Sovereign-AI-Data-Center-Built- With-NVIDIA-Accelerated-Computing.[Accessed 16-05-2025]. Wolf, M. 2025. Japan economic outlook, April 2025. https://w2.deloitte.com/us/en/insights/economy/ asia-pacific/japan-economic-outlook.html. [Accessed 17-05-2025]. Xia, B.; Lu, Q.; Zhu, L.; and Xing, Z. 2024. An AI Sys- tem Evaluation Framework for Advancing AI Safety: Terminology, Taxonomy, Lifecycle Mapping. In Pro- ceedings of the 1st ACM International Conference on AI-Powered Software, 74–78. Yang, Z. 2024. Four things you need to know about China’s AI talent pool. https://w.technologyreview. com/2024/03/27/1090182/ai-talent-global-china-us/. [Accessed 18-05-2025]. YTL. 2022. YTL Corporation Berhad — Latest News — ytl.com.https://w.ytl.com/press-releases/ytl- green-data-center-park-launches-in-johor-the-first- integrated-data-center-park-powered-by-renewable- solar-energy-in-malaysia-2/. [Accessed 16-05-2025]. YTL Group. 2024. Press Releases - YTL AI Cloud — ytlaicloud.com. https://w.ytlaicloud.com/press- releases/ytl-creates-one-of-the-worlds-most- advanced-supercomputers-powered-by-nvidia-grace- blackwell-based-dgx-cloud/. [Accessed 16-05-2025]. Zeb, S.; FNU, N.; Abbasi, N.; and Fahad, M. 2024. AI in Healthcare: Revolutionizing Diagnosis and Ther- apy.https://w.researchgate.net/profile/Nasrullah- Abbasi/publication/383212769 AIinHealthcare RevolutionizingDiagnosisandTherapy/links/ 66c2212e145f4d355366385d/AI-in-Healthcare- Revolutionizing-Diagnosis-and-Therapy.pdf.[Ac- cessed 15-05-2025]. Zeng, Y.; Lu, E.; Guo, X.; Huangfu, C.; Xie, J.; Chen, Y.; Wang, Z.; Liang, D.; Cao, G.; Wang, J.; Ruan, Z.; Guan, X.; and Younas, A. 2025. AI Governance InternationaL Evaluation Index (AGILE Index) 2025. arXiv:2507.11546. Zhou, Q.; and Interesse, G. 2025.China’s Econ- omy in 2024: GDP, Trade, and FDI — china- briefing.com.https://w.china-briefing.com/news/ chinas-economy-in-2024-gdp-trade-fdi/.[Accessed 17-05-2025]. Zuiderwijk, A.; Chen, Y.-C.; and Salem, F. 2021. Im- plications of the Use of Artificial Intelligence in Pub- lic Governance: A Systematic Literature Review and a Research Agenda. Government Information Quarterly, 38(3): 101577. Zulhusni, M. 2025.Alibaba Cloud expands Thai presence with second data centre — cloudcomputing- news.net.https://w.cloudcomputing-news.net/ news/alibaba-cloud-expands-thai-presence-with- second-data-centre/. [Accessed 16-05-2025].