protologue

Chain-of-Thought Monitoring

Also called CoT monitoring, CoT monitorability, reasoning trace monitoring.

Chain-of-thought monitoring is reading a model's reasoning trace, with automated monitors or people, to detect intentions to misbehave, such as plans to cheat on a task, before or alongside its actions.

Description

Because reasoning models often state their intentions in natural language while solving hard tasks, their reasoning offers an unusual view into their decision making. Researchers from several labs have argued that this monitorability is valuable but fragile. Training pressure on the reasoning text, or architectures that reason without words, could remove it, and reasoning traces are not always faithful to the model's actual computation.

Sources

  1. Korbak et al. (2025). Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety.

Cite this entry

Protologue. (2026). Chain-of-Thought Monitoring. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0158). https://protologue.com/t/chain-of-thought-monitoring/

BibTeX
@misc{protologue_chain_of_thought_monitoring,
  title = {Chain-of-Thought Monitoring},
  author = {{Protologue}},
  year = {2026},
  howpublished = {Protologue: A Taxonomy of Prompting and LLM Techniques, v1.4.0},
  note = {Entry PTL-0158},
  url = {https://protologue.com/t/chain-of-thought-monitoring/}
}

Markdown JSON