Microsoft warns AI agents are untrustworthy insider threats

Microsoft launched new security features to protect enterprises from AI agents, including Guardian Agents, which monitors agent behavior as part of Project Perception. Microsoft security chief Hayete Gallot stated that frontier models cannot be trusted, not because they are malicious, but because they are unpredictable and trained to complete tasks rather than follow organizational norms. CEO Satya Nadella argues enterprises must separate the supply of AI intelligence from authority over it. Microsoft also added agent detection to Defender and joined Nvidia's Open Agent Safety Platform, backed by over 100 tech companies.