Chain-of-Thought Monitoring
Chain-of-thought monitoring is reading or automatically checking the reasoning that a model writes out, to detect harmful intentions or misbehaviour before it acts.
What is Chain-of-Thought Monitoring?
Researchers from several labs warned in 2025 that this opportunity may be fragile, because training pressure could make reasoning less faithful or readable.
How it works
A monitor model scans reasoning traces for signs such as plans to cheat or deceive.
← All Artificial Intelligence & Superintelligence definitions · ← Back to the A–Z definitions