mlsanityData to 7 Oct 2026

Chain-of-Thought Monitoring Can Be Unreliable in Implicit-Influence Settings

arXiv 2608.04735 ↗First posted 5 Aug 2026cs.AIcited by 2

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

Abstract (arXiv)

Chain-of-thought (CoT) monitoring is increasingly treated as an important safety layer for frontier reasoning models. Most monitorability evaluations study explicit-influence settings: setups where the prompt directly incentivizes the model to hide something, e.g., by instructing it to perform a hidden side-task. A complementary axis for CoT-monitor evaluations is implicit-influence settings, where the prompt contains no instruction to hide, but the model's behavior is still shaped by features of the task or context, e.g. an irrelevant detail about a candidate that biases a hiring rating. We introduce the first benchmark that directly compares CoT monitorability under the two regimes. We test how model behavior changes in the presence of a nudge to choose a particular option. The nudge is delivered either as a casual aside (implicit), or as a direct instruction to act on the nudge and to conceal having done so (explicit). The benchmark spans four task formats (binary choice, numeric rating, multiple-choice QA, open-ended coding) and seven frontier extended-thinking models. Under explicit influence, a CoT monitor detects 60-94% of behavior shifts: even models instructed to conceal it leak the instruction into their CoT. Under implicit influence, the same factors still shift behavior, but detection falls by 41-46 percentage points in two of our four settings. Realistic system-prompt additions (of the kind a developer might deploy to reduce off-topic bias) lower implicit detection further, to as low as 5%, while preserving the behavioral influence itself. These results suggest that monitorability estimates obtained in explicit-influence settings may over-estimate monitorability, and that monitorability can be further decreased by well-intentioned deployment choices. Our benchmark and code are available at https://github.com/agatha-duzan/implicit-vs-explicit-influence

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 26 of its 35 references that are in the map. First posted on arXiv 5 Aug 2026.

Builds on

26 of 35 references.

Closest ancestors · 14

Same direction, newest first

  1. 2026Noticing the Watcher: LLM Agents Can Infer CoT Monitoring from Blocking Feedback14 Mar
  2. 2026Reasoning Models Struggle to Control their Chains of Thought5 Mar
  3. 2025Monitoring Monitorability20 Dec
  4. 2025Can Reasoning Models Obfuscate Reasoning? Stress-Testing Chain-of-Thought Monitorability21 Oct
  5. 2025LLMs Can Covertly Sandbag on Capability Evaluations Against Chain-of-Thought Monitoring31 Jul
  6. 2025Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety15 Jul
  7. 2025When Chain of Thought is Necessary, Language Models Struggle to Evade Monitors7 Jul
  8. 2025Large language models can learn and generalize steganographic chain-of-thought under process supervision2 Jun
  9. 2025CoT Red-Handed: Stress Testing Chain-of-Thought Monitoring29 May
  10. 2025Reasoning Models Don't Always Say What They Think8 May
  11. 2025Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation14 Mar
  12. 2025Chain-of-Thought Reasoning In The Wild Is Not Always Faithful11 Mar
2 more
  1. 2023Measuring Faithfulness in Chain-of-Thought Reasoning17 Jul
  2. 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting7 May

Other references · 12

  1. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,748
  2. 2024MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmarkcited by 1,758
  3. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,009
  4. 2023Steering Language Models With Activation Engineeringcited by 824
  5. 2025Humanity's Last Examcited by 467
  6. 2025Are DeepSeek R1 And Other Reasoning Models More Faithful?cited by 69
  7. 2025SHADE-Arena: Evaluating Sabotage and Monitoring in LLM Agentscited by 58
  8. 2025Teaching Models to Verbalize Reward Hacking in Chain-of-Thought Reasoningcited by 19
  9. 2026How does information access affect LLM monitors' ability to detect sabotage?cited by 12
  10. 2026Biases in the Blind Spot: Detecting What LLMs Fail to Mentioncited by 8
  11. 2026Censored LLMs as a Natural Testbed for Secret Knowledge Elicitationcited by 5
  12. 2026Value Leakage: An LLM's Answers Are Silently Shaped by Its Own Valuescited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Chain-of-Thought Faithfulness of Reasoning Models Varies with Where and How Preference Cues Are Deliveredcited by 0
  2. 2026Untangling the Mechanisms of Misleading Context in Medical Question Answeringcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Bypassing the Rationale: Causal Auditing of Implicit Reasoning in Language Models
  2. 2025Measuring Chain-of-Thought Monitorability Through Faithfulness and Verbosity
  3. 2026Does Out-of-Sight Equal Out-of-Mind in CoT Monitorability?
  4. 2025A Pragmatic Way to Measure Chain-of-Thought Monitorability
  5. 2026From Decorative to Load-Bearing: Task Difficulty Shapes the Causal Role of Chain-of-Thought
  6. 2026MonitorBench: A Comprehensive Benchmark for Chain-of-Thought Monitorability in Large Language Models
  7. 2026The Fragility of Chain-of-Thought Monitoring Across Typologically Diverse Languages
  8. 2025Investigating CoT Monitorability in Large Reasoning Models