1 paper · 1 filter
Jennifer Za, Julija Bainiaksina, Nikita Ostrovsky +2
Chain-of-thought (CoT) monitoring is a promising safety mechanism for AI agents, based on the premise that visible reasoning traces can surface misaligned or deceptive behavior. Wh…