mlsanityData to 14 Oct 2026

Alignment of LRMs via Counter-Aligned Few-Shot Conversation Exposure

arXiv 2609.27763 ↗First posted 16 Aug 2026cs.AIcited by 0

On the map: LLM reasoning → Reasoning for agents, planning and code

also in 1 more
Abstract (arXiv)

Large Reasoning Models (LRMs) rely on explicit chain-of-thought (CoT) reasoning and large context windows to achieve strong performance on complex tasks, but these features also introduce new attack surfaces. We show that LRMs' reasoning processes can be systematically steered by prepending counter-aligned few-shot conversations containing explicit CoT traces, leading to unsafe generations on harmful queries and unwarranted refusals on benign ones. We formalize this attack as SRCF (Steering Reasoning via Counter-Aligned Few-shot Conversations) that operates solely through a flexible conversational interface and requires no access to the model's parameters and gradients. Our key insight is that SRCF exploits an adversarial generalization issue that induces a representation drift, causing the representations of benign and harmful inputs to shift in a similar direction. This observation motivates our post-training defense, ARCF (Aligning Reasoning via Counter-Aligned Few-Shot Conversations), which exposes models to counter-aligned conversational contexts while enforcing aligned targets. ARCF is compatible with existing post-training methods and consistently improves safety and helpfulness without degrading utility.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 53 of its 60 references that are in the map. First posted on arXiv 16 Aug 2026.

Builds on

53 of 60 references.

Closest ancestors · 23

Same direction, newest first

  1. 2026Conflicts Make Large Reasoning Models Vulnerable to Attacks10 Apr
  2. 2026How Does the Thinking Step Influence Model Safety? An Entropy-based Safety Reminder for LRMs7 Jan
  3. 2025Beyond SFT: Reinforcement Learning for Safer Large Reasoning Models with Better Reasoning Ability1 Dec
  4. 2025Chain-of-Thought Hijacking30 Oct
  5. 2025Large Reasoning Models Learn Better Alignment from Flawed Thinking1 Oct
  6. 2025AdvChain: Adversarial Chain-of-Thought Tuning for Robust Safety Alignment of Large Reasoning Models29 Sep
  7. 2025Three Minds, One Legend: Jailbreak Large Reasoning Model with Adaptive Stacked Ciphers22 May
  8. 2025SAFEPATH: Preventing Harmful Reasoning in Chain-of-Thought via Early Alignment20 May
  9. 2025FalseReject: A Resource for Improving Contextual Safety and Mitigating Over-Refusals in LLMs via Structured Reasoning12 May
  10. 2025Safety in Large Reasoning Models: A Survey24 Apr
  11. 2025RealSafe-R1: Safety-Aligned DeepSeek-R1 without Compromising Reasoning Capability14 Apr
  12. 2025ShadowCoT: Cognitive Hijacking for Stealthy Reasoning Backdoors in LLMs8 Apr
11 more
  1. 2025STAR-1: Safer Alignment of Reasoning LLMs with 1K Data2 Apr
  2. 2025Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonable1 Mar
  3. 2025A Mousetrap: Fooling Large Reasoning Models for Jailbreak with Chain of Iterative Chaos19 Feb
  4. 2025H-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 Thinking18 Feb
  5. 2025The Hidden Risks of Large Reasoning Models: A Safety Assessment of R118 Feb
  6. 2025SafeChain: Safety of Language Models with Long Chain-of-Thought Reasoning Capabilities17 Feb
  7. 2025Reasoning-Augmented Conversation for Multi-Turn Jailbreak Attacks on Large Language Models16 Feb
  8. 2025OverThink: Slowdown Attacks on Reasoning LLMs4 Feb
  9. 2025Trading Inference-Time Compute for Adversarial Robustness31 Jan
  10. 2025DarkMind: Latent Chain-of-Thought Backdoor in Customized LLMs24 Jan
  11. 2024BadChain: Backdoor Chain-of-Thought Prompting for Large Language Models20 Jan

Other references · 30

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  3. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  4. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  5. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
  6. 2024MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmarkcited by 1,848
  7. 2025gpt-oss-120b & gpt-oss-20b Model Cardcited by 1,386
  8. 2021A General Language Assistant as a Laboratory for Alignmentcited by 1,055
  9. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  10. 2024A Survey on Large Language Models for Code Generationcited by 658
  11. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMscited by 570
  12. 2023XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Modelscited by 505
18 more
  1. 2023Jailbreak and Guard Aligned Language Models with Only Few In-Context Demonstrationscited by 431
  2. 2025Reasoning Models Don't Always Say What They Thinkcited by 415
  3. 2021Recursively Summarizing Books with Human Feedbackcited by 290
  4. 2024OR-Bench: An Over-Refusal Benchmark for Large Language Modelscited by 244
  5. 2025Qwen3Guard Technical Reportcited by 155
  6. 2025The Danger of Overthinking: Examining the Reasoning-Action Dilemma in Agentic Taskscited by 152
  7. 2024Safety Layers in Aligned Large Language Models: The Key to LLM Securitycited by 134
  8. 2023Adversarial Demonstration Attacks on Large Language Modelscited by 63
  9. 2025Blending Supervised and Reinforcement Fine-Tuning with Prefix Samplingcited by 60
  10. 2024Large Language Model Safety: A Holistic Surveycited by 59
  11. 2023Hijacking Large Language Models via Adversarial In-Context Learningcited by 45
  12. 2024Tradeoffs Between Alignment and Helpfulness in Language Models with Steering Methodscited by 28
  13. 2025FORTRESS: Frontier Risk Evaluation for National Security and Public Safetycited by 24
  14. 2025Innate Reasoning is Not Enough: In-Context Learning Enhances Reasoning Large Language Models with Less Overthinkingcited by 20
  15. 2025Multi-Turn Jailbreaks Are Simpler Than They Seemcited by 17
  16. 2025Discovering Forbidden Topics in Language Modelscited by 9
  17. 2025Self-Correcting Large Language Models: Generation vs. Multiple Choicecited by 6
  18. 2025Thinking with Nothinking Calibration: A New In-Context Learning Paradigm in Reasoning Large Language Modelscited by 4

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Reasoning Hijacking: The Fragility of Reasoning Alignment in Large Language Models
  2. 2026Does Deeper Reasoning Compromise Alignment? Revealing and Mitigating of Alignment Collapse in Large Reasoning Models
  3. 2025Distractor Injection Attacks on Large Reasoning Models: Characterization and Defense
  4. 2025Refining Input Guardrails: Enhancing LLM-as-a-Judge Efficiency Through Chain-of-Thought Fine-Tuning and Alignment
  5. 2025Eliciting and Analyzing Emergent Misalignment in State-of-the-Art Large Language Models
  6. 2025ENIGMA: The Geometry of Reasoning and Alignment in Large-Language Models
  7. 2025A Survey on Post-training of Large Language Models
  8. 2026On the Efficiency-Safety Dilemma in Large Reasoning Models