Why Safety Probes Catch Liars But Miss Fanatics
Activation-based probes have emerged as a promising approach for detecting deceptively aligned AI systems by identifying internal conflict between true and stated goals. We identify a fundamental blind spot: probes fail on coherent misalignment - models that believe their harmful behavior is virtuous rather than strategically hiding it. We prove that no polynomial-time probe can detect such misalignment with non-trivial accuracy when belief structures reach sufficient complexity (PRF-like triggers). We show the emergence of this phenomenon on a simple task by training two models with identical RLHF procedures: one producing direct hostile responses ("the Liar"), another trained towards coherent misalignment using rationalizations that frame hostility as protective ("the Fanatic"). Both exhibit identical behavior, but the Liar is detected 95%+ of the time while the Fanatic evades detection almost entirely. We term this Emergent Probe Evasion: training with belief-consistent reasoning shifts models from a detectable "deceptive" regime to an undetectable "coherent" regime - not by learning to hide, but by learning to believe.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
Before the Last Token: Diagnosing Final-Token Safety Probe Failures
Final-token safety probes monitor a single hidden state after prompt prefill, but jailbreak prompts can contain probe-visible unsafe evidence distributed across earlier user-token representations that is missed by this r…
Liars' Bench: Evaluating Lie Detectors for Language Models
Prior work has introduced techniques for detecting when large language models (LLMs) lie, that is, generate statements they believe are false. However, these techniques are typically validated in narrow settings that do …
One Probe Won't Catch Them All: Towards Targeted Deception Detection
Linear probes are a promising approach for monitoring AI systems for deceptive behaviour. Previous work has shown that a linear classifier trained on a contrastive instruction pair and a simple dataset can achieve good p…
Beyond Linear Probes: Dynamic Safety Monitoring for Language Models
Monitoring large language models' (LLMs) activations is an effective way to detect harmful requests before they lead to unsafe outputs. However, traditional safety monitors often require the same amount of compute for ev…
Catching rationalization in the act: detecting motivated reasoning before and after CoT via activation probing
Large language models (LLMs) can produce chains of thought (CoT) that do not accurately reflect the actual factors driving their answers. In multiple-choice settings with an injected hint favoring a particular option, mo…