← Back to papers

Paper deep dive

CoT Defender: Preemptive Chain-of-Thought Occupation for Jailbreak Attack Mitigation

Xiaokang Li, Jin Liu, Yongqiang Tang, Zhiwen Xie, Yihe Wang, Xiao Yu, Long Zhao, Bo Huang

Year: 2026Venue: Neural NetworksArea: Adversarial RobustnessType: EmpiricalEmbeddings: 0

Models: unspecified four LLMs

Abstract

CoT Defender preemptively occupies LLM initial tokens with chain-of-thought reasoning using a two-stage RL training framework, reducing jailbreak attack success rates below 8% across four models.

Tags

adversarial-robustness (suggested, 80%)ai-safety (imported, 100%)empirical (suggested, 88%)

Links

Intelligence

Status: not_run | Model: - | Prompt: - | Confidence: 0%

Entities (0)

No extracted entities yet.

Relation Signals (0)

No relation signals yet.

Cypher Suggestions (0)

No Cypher suggestions yet.

Full Text

No full-text extraction is stored for this paper yet.