"Did you lie?" Evaluating Lie Detectors across Model Scale and Belief-Verified Model Organisms

Researchers have developed new testbeds to evaluate lie detectors for language models, finding that existing detection methods often fail when models are trained to hold false beliefs. The study tested four detectors across 31 models, revealing that only a chain-of-thought judge maintained strong accuracy at 0.82, while activation- and logprob-based detectors dropped sharply on trained model organisms. The findings suggest current lie detectors cannot support high-confidence claims about model beliefs, limiting their use for auditing and monitoring AI behavior.

arXiv:2606.12618v1 Announce Type: new Abstract: Robust lie detectors for language models could enable powerful techniques for auditing, monitoring, and post-hoc investigation of model behaviour, but evaluating them requires testbeds where models verifiably believe the opposite of what they say. We show that existing trained model organisms often fail this requirement, leaving prior positive and negative detection results difficult to interpret. We address this with 13 reasoning model organisms whose hidden beliefs are verified in chain-of-thought and shown to generalise to held-out tasks, alongside Varied Deception, a prompted-lying testbed covering a broad range of lie-inducing motivations. On these testbeds we evaluate four detectors: a chain-of-thought judge, a logprob classifier, and two activation probes, including Did-You-Lie DYL , a new method for training follow-up probes. On prompted lying, across 31 open-weight models spanning 2B to 1T parameters, all four detectors show positive scaling with model capability. However, every activation- and logprob-based detector drops sharply on our trained model organisms, with DYL retaining the most signal; only the chain-of-thought judge remains strong, with 0.82 balanced accuracy, partly as an artefact of our verification process favouring CoT-readable beliefs. Current lie detectors therefore cannot support high-confidence claims about model beliefs, and we suggest research directions that may address some of their current limitations. We release our datasets, model organisms, and trained detectors.