mlsanityData to 7 Oct 2026

A Concrete Roadmap towards Safety Cases based on Chain-of-Thought Monitoring

arXiv 2510.19476 ↗First posted 22 Oct 2025cs.LG cs.AIcited by 2

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

Abstract (arXiv)

As AI systems approach dangerous capability levels where inability safety cases become insufficient, we need alternative approaches to ensure safety. This paper presents a roadmap for constructing safety cases based on chain-of-thought (CoT) monitoring in reasoning models and outlines our research agenda. We argue that CoT monitoring might support both control and trustworthiness safety cases. We propose a two-part safety case: (1) establishing that models lack dangerous capabilities when operating without their CoT, and (2) ensuring that any dangerous capabilities enabled by a CoT are detectable by CoT monitoring. We systematically examine two threats to monitorability: neuralese and encoded reasoning, which we categorize into three forms (linguistic drift, steganography, and alien reasoning) and analyze their potential drivers. We evaluate existing and novel techniques for maintaining CoT faithfulness. For cases where models produce non-monitorable reasoning, we explore the possibility of extracting a monitorable CoT from a non-monitorable CoT. To assess the viability of CoT monitoring safety cases, we establish prediction markets to aggregate forecasts on key technical milestones influencing their feasibility.

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning. Builds on 34 of its 54 references that are in the map. First posted on arXiv 22 Oct 2025.

Builds on

34 of 54 references.

Closest ancestors · 13

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. 2025Thought Anchors: Which LLM Reasoning Steps Matter?23 Jun
  4. 2025Large language models can learn and generalize steganographic chain-of-thought under process supervision2 Jun
  5. 2025CoT Red-Handed: Stress Testing Chain-of-Thought Monitoring29 May
  6. 2025Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation14 Mar
  7. 2025Chain-of-Thought Reasoning In The Wild Is Not Always Faithful11 Mar
  8. 2024Training Large Language Models to Reason in a Continuous Latent Space9 Dec
  9. 2024Let's Think Dot by Dot: Hidden Computation in Transformer Language Models24 Apr
  10. 2024Chain of Thought Empowers Transformers to Solve Inherently Serial Problems20 Feb
  11. 2023Preventing Language Models From Hiding Their Reasoning27 Oct
  12. 2023Measuring Faithfulness in Chain-of-Thought Reasoning17 Jul
1 more
  1. 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting7 May

Other references · 21

  1. 2017Attention Is All You Needcited by 59,932
  2. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,491
  3. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  4. 2017Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithmcited by 723
  5. 2024Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Trainingcited by 478
  6. 2025Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approachcited by 372
  7. 2024Alignment faking in large language modelscited by 316
  8. 2024Frontier Models are Capable of In-context Schemingcited by 252
  9. 2025ARC-AGI-2: A New Challenge for Frontier AI Reasoning Systemscited by 171
  10. 2023Sociotechnical Safety Evaluation of Generative AI Systemscited by 154
  11. 2025Agentic Misalignment: How LLMs Could Be Insider Threatscited by 136
  12. 2024AI Sandbagging: Language Models can Strategically Underperform on Evaluationscited by 131
9 more
  1. 2024Evaluating Frontier Models for Dangerous Capabilitiescited by 121
  2. 2025Large Language Models Often Know When They Are Being Evaluatedcited by 105
  3. 2025Stress Testing Deliberative Alignment for Anti-Scheming Trainingcited by 71
  4. 2025Understanding Reasoning in Thinking Language Models via Steering Vectorscited by 69
  5. 2024Hidden in Plain Text: Emergence & Mitigation of Steganographic Collusion in LLMscited by 52
  6. 2024Safety Cases: How to Justify the Safety of Advanced AI Systemscited by 48
  7. 2025A sketch of an AI control safety casecited by 36
  8. 2025Early Signs of Steganographic Capabilities in Frontier LLMscited by 19
  9. 2024Frontier AI systems have surpassed the self-replicating red linecited by 16

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Monitor Jailbreaking: Evading Chain-of-Thought Monitoring Without Encoded Reasoningcited by 0
  2. 2026Learning Steganography Is Easy, Learning Steganographic Reasoning Is Hardcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Investigating CoT Monitorability in Large Reasoning Models
  2. 2025A Pragmatic Way to Measure Chain-of-Thought Monitorability
  3. 2025Monitoring Monitorability
  4. 2026Evading Chain-of-Thought Monitoring Through Model Poisoning
  5. 2026Towards Safer Large Reasoning Models by Promoting Safety Decision-Making before Chain-of-Thought Generation
  6. 2025Reasoning Models Don't Always Say What They Think
  7. 2026How Does the Thinking Step Influence Model Safety? An Entropy-based Safety Reminder for LRMs
  8. 2025Annotating the Chain-of-Thought: A Behavior-Labeled Dataset for AI Safety