Microsoft launches agent security feature

Microsoft introduced Guardian Agents on Monday, a new security feature designed to monitor the AI agents that enterprises deploy and use.

Guardian Agents builds on Project Perception

Guardian Agents is part of Project Perception, a product Microsoft introduced in June with red, blue and green teams to test and fix AI vulnerabilities.

Microsoft shifts focus to internal threats

Project Perception originally assumed AI hackers would come from outside companies. But Microsoft CEO Satya Nadella now warns that deployed AI agents themselves pose insider threat risks.

Nadella: stop trusting AI model providers

In his private blog earlier this month, Nadella warned enterprises to stop trusting AI model providers and implement separate systems to watch and control agents.

“We need to separate the supply of intelligence from the authority over it.”

Frontier models cannot be trusted

Microsoft security chief Hayete Gallot stated frontier models cannot be trusted because they are unpredictable and trained to achieve tasks, not follow organizational norms.

“Unfortunately, frontier models, regardless of their origin, cannot be trusted.”

Microsoft's broader security strategy

Beyond Guardian Agents, Microsoft added agent detection to Microsoft Defender, features limiting agent data access, and dynamic permission adjustments requiring human approval for risky decisions.

Industry adopts agent safety standards

Microsoft joined Nvidia's Open Agent Safety Platform, supported by more than one hundred other tech companies, which isolates agents and can monitor and stop behavior through hardware.

Microsoft launches agent-focused hardware

Microsoft announced new Surface Laptop Ultra PCs built for agents, running on Nvidia chips with agent security and controls built in to the hardware.