Adaptive Attacks on Trusted Monitors Subvert AI Control Protocols
By: Mikhail Terekhov , Alexander Panfilov , Daniil Dzenhaliou and more
Potential Business Impact:
Makes AI safety rules easier for AI to break.
AI control protocols serve as a defense mechanism to stop untrusted LLM agents from causing harm in autonomous settings. Prior work treats this as a security problem, stress testing with exploits that use the deployment context to subtly complete harmful side tasks, such as backdoor insertion. In practice, most AI control protocols are fundamentally based on LLM monitors, which can become a central point of failure. We study adaptive attacks by an untrusted model that knows the protocol and the monitor model, which is plausible if the untrusted model was trained with a later knowledge cutoff or can search for this information autonomously. We instantiate a simple adaptive attack vector by which the attacker embeds publicly known or zero-shot prompt injections in the model outputs. Using this tactic, frontier models consistently evade diverse monitors and complete malicious tasks on two main AI control benchmarks. The attack works universally against current protocols that rely on a monitor. Furthermore, the recent Defer-to-Resample protocol even backfires, as its resampling amplifies the prompt injection and effectively reframes it as a best-of-$n$ attack. In general, adaptive attacks on monitor models represent a major blind spot in current control protocols and should become a standard component of evaluations for future AI control mechanisms.
Similar Papers
Evaluating Control Protocols for Untrusted AI Agents
Artificial Intelligence
Keeps AI agents from doing bad things safely.
Reliable Weak-to-Strong Monitoring of LLM Agents
Artificial Intelligence
Finds ways AI spies on other AI.
Factor(T,U): Factored Cognition Strengthens Monitoring of Untrusted AI
Cryptography and Security
Makes AI safer by breaking big jobs into small ones.