arXiv cs.AI
  • Score 92
  • Official

Chain-of-Thought Controllability Can Reduce Chain-of-Thought Monitorability

arXiv:2611.21799v1 Announce Type: new Abstract: Chain-of-thought (CoT) monitoring is a promising tool for detecting hidden goals, off-task behavior, and other failures in AI agents. One risk is that models could deliberately shape their reasoning to make it opaque to monitors. We stress-test this risk through persona-control training: supervised fine-tuning on mathematical reasoning traces rewritt

So what

What this event means by reading role—not a longer recap.

  • BuilderIf you rely on CoT monitoring for agent safety, persona-style fine-tuning can silently degrade it; transparency prompts may partially recover monitorability.
  • ResearcherShows controllability training generalizes to monitor evasion without evasion examples, and offers model organisms plus transparency-prompt mitigations as a tes
  • InvestorUndermines CoT monitoring as a cheap safety layer for agent products, raising compliance and trust costs for agentic deployments.

Score dimensions

Higher total means read first. Each bar is one factor we use to rank the system pool. How we score

RelevanceHow tightly this is about AI.
88
ImpactHow much this could change the field or the market.
72
NoveltyHow new this is versus a recap.
80
CredibilityHow much we trust the source.
97
ActionabilityWhether a reader can do something with it.
55