Caught in the Act: Probes Effectively Detect Sabotage and Catch Unverbalized Deception
AI lie detectors that catch language models when they hide the truth
Researchers built software probes that detect when large language models deliberately deceive users, achieving 98.8% accuracy on deception detection tests. The probes work even when a model's lies would be undetectable from conversation alone—finding hidden goals and false claims buried in the model's internal processing, including false statements about politically sensitive topics.
As AI systems take on higher-stakes roles, we need reliable ways to catch deception before it reaches users. These probes could enable safety teams to deploy monitoring systems that catch lying during live operation, not just in hindsight, reducing the window where a model can mislead people without detection. The authors released their training dataset to help the field scale this capability across different models.