Pith. sign in

REVIEW 15 cited by

H-CoT: Hijacking the Chain-of-Thought Safety Reasoning Mechanism to Jailbreak Large Reasoning Models, Including OpenAI o1/o3, DeepSeek-R1, and Gemini 2.0 Flash Thinking

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2502.12893 v2 pith:LYBZL6E5 submitted 2025-02-18 cs.CL

classification cs.CL
keywords reasoningsafetymodelchain-of-thoughtdeepseek-r1flashgeminih-cot
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Large Reasoning Models (LRMs) have recently extended their powerful reasoning capabilities to safety checks-using chain-of-thought reasoning to decide whether a request should be answered. While this new approach offers a promising route for balancing model utility and safety, its robustness remains underexplored. To address this gap, we introduce Malicious-Educator, a benchmark that disguises extremely dangerous or malicious requests beneath seemingly legitimate educational prompts. Our experiments reveal severe security flaws in popular commercial-grade LRMs, including OpenAI o1/o3, DeepSeek-R1, and Gemini 2.0 Flash Thinking. For instance, although OpenAI's o1 model initially maintains a high refusal rate of about 98%, subsequent model updates significantly compromise its safety; and attackers can easily extract criminal strategies from DeepSeek-R1 and Gemini 2.0 Flash Thinking without any additional tricks. To further highlight these vulnerabilities, we propose Hijacking Chain-of-Thought (H-CoT), a universal and transferable attack method that leverages the model's own displayed intermediate reasoning to jailbreak its safety reasoning mechanism. Under H-CoT, refusal rates sharply decline-dropping from 98% to below 2%-and, in some instances, even transform initially cautious tones into ones that are willing to provide harmful content. We hope these findings underscore the urgent need for more robust safety mechanisms to preserve the benefits of advanced reasoning capabilities without compromising ethical standards.

Discussion (0). Sign in to comment.

Forward citations

Cited by 15 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. TRAP: Hijacking VLA CoT-Reasoning via Adversarial Patches

    cs.CR 2026-03 conditional novelty 7.5 of 10

    Physical adversarial patches can steer CoT-reasoning VLAs into attacker-specified manipulation behaviors without changing the user’s instruction.

  2. Unable to Forget: Proactive Interference Reveals Working Memory Limits in LLMs Beyond Context Length

    cs.CL 2025-06 conditional novelty 7.0 of 10

    LLM retrieval accuracy decays log-linearly with the number of interfering prior updates, and 'forget' instructions barely improve it.

  3. CoreMatching: A Co-adaptive Sparse Inference Framework with Token and Neuron Pruning for Comprehensive Acceleration of Vision-Language Models

    cs.LG 2025-05 conditional novelty 7.0 of 10

    CoreMatching couples token pruning and neuron pruning in vision-language models by selecting tokens that activate the most core neurons, achieving large inference speedups with minor accuracy loss.

  4. TYPO: Instruction-Dense Visual Jailbreaks against Commercial Closed-Source Image-Generation Models

    cs.CR 2026-07 conditional novelty 6.5 of 10

    Safety alignment fails to transfer from text replies to text-in-image, and TYPO’s dual-channel combinatorial search jailbreaks four commercial image models at >90% ASR for ~$0.04.

  5. Chain-of-Models: Cross-Model Auditing for Bias-Robust LLM Judges

    cs.CL 2026-05 conditional novelty 6.0 of 10

    Per-bias selection of a cross-family LLM auditor lifts biased-judgment accuracy from 0.805/0.824 baselines to 0.884.

  6. Safety in Embodied AI: A Survey of Risks, Attacks, and Defenses

    cs.CR 2026-03 unverdicted novelty 6.0 of 10

    The survey organizes over 400 papers on embodied AI safety into a multi-level taxonomy and flags overlooked issues such as fragile multimodal fusion and unstable planning under jailbreaks.

  7. Does More Inference-Time Compute Really Help Robustness?

    cs.AI 2025-07 conditional novelty 6.0 of 10

    With exposed reasoning chains, increasing inference-time compute consistently decreases measured robustness across 12 open-source reasoning models, while hidden chains show improvements.

  8. The Other Mind: How Language Models Exhibit Human Temporal Cognition

    cs.AI 2025-07 conditional novelty 6.0 of 10

    Larger LLMs develop a subjective 'present' around the current date, and their year similarity judgments follow a logarithmic Weber-Fechner compression, with supporting neural and representational evidence.

  9. Beyond Safe Answers: A Benchmark for Evaluating True Risk Awareness in Large Reasoning Models

    cs.AI 2025-05 conditional novelty 6.0 of 10

    A new benchmark shows that top reasoning models identify all relevant risks in under 40% of cases even when their final answers look safe.

  10. SafeTy Reasoning Elicitation Alignment for Multi-Turn Dialogues

    cs.CL 2025-05 conditional novelty 5.0 of 10

    STREAM fine-tunes a small reasoning model on human-labeled, reason-annotated multi-turn dialogues and uses it to warn target LLMs, cutting average attack success rates by roughly half while keeping benchmark scores close.

  11. Reasoning Can Hurt the Inductive Abilities of Large Language Models

    cs.CV 2025-05 conditional novelty 5.0 of 10

    Chain-of-thought reasoning can hurt LLMs' ability to infer hidden rules from gameplay transcripts, and structured interventions recover the lost accuracy.

  12. A Multimodal Automatic Redteaming Evaluation based on Atomic Jailbreak Strategy Decoupling and Combination

    cs.CR 2026-08 conditional novelty 4.0 of 10

    A new jailbreak framework, HACA, combines atomic text and image attack strategies selected by a cross-modal planner and generates attacks with LLMs and text-to-image models, reaching 95.48% average attack success acro...

  13. A Comprehensive Survey on Trustworthiness in Reasoning with Large Language Models

    cs.CL 2025-09 conditional novelty 4.0 of 10

    A structured literature survey concluding that reasoning capabilities do not automatically make LLMs more trustworthy and can introduce new vulnerabilities in safety, robustness, and privacy.

  14. SoK: A Comprehensive Security Analysis of Jailbreak Resilience in GPT and DeepSeek Models

    cs.CR 2025-06 conditional novelty 4.0 of 10

    Across 510 HarmBench behaviors and seven attack methods, GPT-4 models show more consistent jailbreak resilience than DeepSeek models, whose vulnerability grows with scale.

  15. A Red Teaming Roadmap Towards System-Level Safety

    cs.CR 2025-05 conditional novelty 4.0 of 10

    A position paper from Scale AI argues that red teaming research should prioritize product-level safety specifications, realistic attacker models, and system-level monitoring over abstract model-level harm benchmarks.

Pith tools