Microsoft researchers demonstrate hidden backdoor technique for AI agents

Microsoft researchers demonstrated a backdoor technique called Lazarus that implants a hidden trigger in an AI model through small changes, causing an agent to execute an attacker's command when the trigger appears, Help Net Security reported on November 11, 2026. The technique requires no model retraining and lets the agent behave normally during standard safety evaluations. Microsoft said it has not observed threat actors weaponizing Lazarus but warned the research highlights a security concern for organizations that give AI agents access to terminals.