Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents - Microsoft Research

Microsoft researchers have discovered a backdoor technique called Lazarus, which can reprogram a small number of dormant neurons in an AI model without retraining. The method allows an AI agent to execute malicious commands when a specific trigger appears while still passing standard safety evaluations. Researchers emphasize that weight integrity and model provenance are critical to defending against this type of threat. Open article →