Gerard Boxo | Searcharxiv

arXiv2025

Linear probes rely on textual evidence: Results from leakage mitigation studies in language models

White-box monitors are a popular technique for detecting potentially harmful behaviours in language models. While they perform well in general, their effectiveness in detecting text-ambiguous behaviour is disputed. In this work, we find evidence that removing textual evidence of a behaviour significantly decreases probe performance. The AUROC reduction ranges from $10$- to $30$-point depending on the setting. We evaluate probe monitors across three setups (Sandbagging, Sycophancy, and Bias), finding that when probes rely on textual evidence of the target behaviour (such as system prompts or CoT reasoning), performance degrades once these tokens are filtered. This filtering procedure is standard practice for output monitor evaluation. As further evidence of this phenomenon, we train Model Organisms which produce outputs without any behaviour verbalisations. We validate that probe performance on Model Organisms is substantially lower than unfiltered evaluations: $0.57$ vs $0.74$ AUROC for Bias, and $0.57$ vs $0.94$ AUROC for Sandbagging. Our findings suggest that linear probes may be brittle in scenarios where they must detect non-surface-level patterns.

cs.AI↗

arXiv2025

Caught in the Act: a mechanistic approach to detecting deception

Sophisticated instrumentation for AI systems might have indicators that signal misalignment from human values, not unlike a "check engine" light in cars. One such indicator of misalignment is deceptiveness in generated responses. Future AI instrumentation may have the ability to detect when an LLM generates deceptive responses while reasoning about seemingly plausible but incorrect answers to factual questions. In this work, we demonstrate that linear probes on LLMs internal activations can detect deception in their responses with extremely high accuracy. Our probes reach a maximum of greater than 90% accuracy in distinguishing between deceptive and non-deceptive arguments generated by llama and qwen models ranging from 1.5B to 14B parameters, including their DeepSeek-r1 finetuned variants. We observe that probes on smaller models (1.5B) achieve chance accuracy at detecting deception, while larger models (greater than 7B) reach 70-80%, with their reasoning counterparts exceeding 90%. The layer-wise probe accuracy follows a three-stage pattern across layers: near-random (50%) in early layers, peaking in middle layers, and slightly declining in later layers. Furthermore, using an iterative null space projection approach, we find multitudes of linear directions that encode deception, ranging from 20 in Qwen 3B to nearly 100 in DeepSeek 7B and Qwen 14B models.

cs.AI↗