MICROSOFT SAYS AI AGENTS ARE LIKE INSIDER THREATS AS FRONTIER MODELS CANT BE TRUSTED

What happened: Microsoft introduced new security features, including Guardian Agents, to monitor AI agents used by enterprises. The feature is part of Project Perception, a product introduced in June. Key numbers: Project Perception launched three agent classes. Microsoft Defender can detect up to 60 types of agents. Why it matters: Microsoft executives warn that frontier models cannot be trusted because they are unpredictable, and Nadella argues enterprises must separate the supply of intelligence from authority over it.

Microsoft introduced new security features, including Guardian Agents, to monitor AI agents used by enterprises. The feature is part of Project Perception, a product introduced in June.

Microsoft executives warn that frontier models cannot be trusted because they are unpredictable, and Nadella argues enterprises must separate the supply of intelligence from authority over it.