mlsanityData to 14 Oct 2026

Output Supervision Can Obfuscate the Chain of Thought

arXiv 2511.11584 ↗First posted 11 Oct 2025cs.LG cs.AI cs.CRcited by 13

On the map: LLM reasoning → Prompting and chain-of-thought reasoning

Abstract (arXiv)

OpenAI (2025) showed that training against a chain of thought (CoT) monitor can cause obfuscated CoTs, which contain bad behavior the monitor cannot detect. They proposed to keep CoTs monitorable by training only against output monitors that do not have access to CoT. We show that such training can still cause obfuscated CoTs via two mechanisms. First, when a model is trained to produce a safe-looking output, that model may generalize to making its CoTs look safe. Second, since later tokens are conditioned on earlier ones, safe-looking CoTs may increase the likelihood of safe outputs, causing safe-looking CoTs to be reinforced. We introduce two mitigations to address these two issues, which achieve a Pareto improvement in terms of monitorability and task performance compared to regular training.

Key facts. Cited by 13 papers in the map (data through 14 Oct 2026). Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning. Builds on 10 of its 43 references that are in the map. First posted on arXiv 11 Oct 2025.

Builds on

10 of 43 references.

Closest ancestors · 5

Same direction, newest first

  1. 2025Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety15 Jul
  2. 2025When Chain of Thought is Necessary, Language Models Struggle to Evade Monitors7 Jul
  3. 2025Reasoning Models Don't Always Say What They Think8 May
  4. 2025Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation14 Mar
  5. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Models28 Jan

Other references · 5

  1. 2025Qwen3 Technical Reportcited by 8,820
  2. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  3. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  4. 2025REASONING GYM: Reasoning Environments for Reinforcement Learning with Verifiable Rewardscited by 130
  5. 2024Safety Cases: How to Justify the Safety of Advanced AI Systemscited by 48

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 13.

13 papers.

Sort:

By area:

  1. 2026Reasoning Models Struggle to Control their Chains of Thoughtcited by 25
  2. 2026Chain-of-thought obfuscation learned from output supervision can generalise to unseen taskscited by 5
  3. 2026Consistency Training while Mitigating Obfuscation via Rate Matchingcited by 4
  4. 2026Multi-Stream LLMs: Unblocking Language Models with Parallel Streams of Thoughts, Inputs and Outputscited by 3
  5. 2026Training on Documents About Monitoring Leads to CoT Obfuscationcited by 3
  6. 2026Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluationscited by 2
  7. 2026Why Safety Probes Catch Liars But Miss Fanaticscited by 1
  8. 2026HarmThoughts: A Benchmark for Fine-Grained Harmful Behavior Detection in Reasoning Tracescited by 1
  9. 2026A transformer architecture alteration to incentivise externalised reasoningcited by 0
  10. 2026Tandem Reinforcement Learning with Verifiable Rewardscited by 0
  11. 2026On the Generalization of Steering Vectors for Chain-of-Thought Faithfulnesscited by 0
  12. 2026Taming CoT Obfuscation in VLMs: From Mechanistic Evidence to Activation Enforcementcited by 0
1 more
  1. 2026Monitorability Disposition in Large Reasoning Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Can Reasoning Models Obfuscate Reasoning? Stress-Testing Chain-of-Thought Monitorability
  2. 2025Large language models can learn and generalize steganographic chain-of-thought under process supervision
  3. 2026Monitor Jailbreaking: Evading Chain-of-Thought Monitoring Without Encoded Reasoning
  4. 2026Evading Chain-of-Thought Monitoring Through Model Poisoning
  5. 2026Corrupt Plans, Clean Traces: Evading Chain-of-Thought Monitoring with Plan Injection
  6. 2025CoT Red-Handed: Stress Testing Chain-of-Thought Monitoring
  7. 2026Analyzing and Improving Chain-of-Thought Monitorability Through Information Theory
  8. 2026Chain of Thought Monitoring Can Fail Even When CoT Is Necessary