Bootstrapped Monitoring: Leveraging Transparent Reasoning to Oversee Stronger AI Agents
For AI safety researchers, this provides a practical protocol to extend the useful lifetime of trusted oversight models as frontier AI capabilities grow.
Bootstrapped monitoring addresses the problem of trusted models becoming unreliable monitors as AI capabilities advance. The protocol uses a stronger untrusted model with transparent chain-of-thought reasoning overseen by a weaker trusted model, achieving substantially improved catch rates on multi-turn software engineering tasks even under active collusion.
Trusted monitoring is a cornerstone of AI control. However, as frontier models grow more capable, the increasing capabilities gap between trusted and untrusted models may render trusted models unreliable monitors. We introduce \emph{bootstrapped monitoring}, a protocol that addresses this by inserting a stronger, intermediate untrusted model with transparent chain-of-thought reasoning into the oversight chain. The untrusted monitor ($U_m$) evaluates the agent's actions, while a weaker trusted model ($T$) oversees $U_m$'s reasoning to detect collusion. We evaluate bootstrapped monitoring on multi-turn software engineering tasks (BashArena) across multiple agents and monitors. Bootstrapped monitoring substantially improves catch rates over trusted-only monitoring, even when the untrusted monitor actively colludes with the agent, provided we have access to its raw chain-of-thought. Our results suggest that bootstrapped monitoring can extend the useful lifetime of trusted models in control as AI capabilities advance.