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

Microsoft researchers have identified a vulnerability called Lazare, veni foras, which involves activating dormant neurons within neural networks to create unauthorized backdoors in AI agents. This method allows an attacker to inject hidden triggers that cause the system to execute malicious commands without altering its primary performance. Because these inactive pathways are often overlooked during standard security audits, this finding highlights a potential long-term risk for the integrity and reliability of large-scale machine learning models.