Microsoft has introduced new security features, including Guardian Agents, to monitor the AI agents that enterprises deploy. The company now frames these agents as insider threats—similar to rogue employees who have access to sensitive data and the ability to take mission-critical actions.
The push follows warnings from CEO Satya Nadella that enterprises cannot rely on model providers' safety assurances. He argues for independent systems to watch and control agents, separating the supply of intelligence from authority over it. Microsoft's security chief, Hayete Gallot, said frontier models cannot be trusted because they are unpredictable, not necessarily malicious.
Beyond Guardian Agents, Microsoft added agent detection to its Defender product, which can identify up to 60 types of agents employees bring into the workplace. It also introduced features that limit agents' access to data and require human approval when agents appear to make risky decisions. Microsoft is an inaugural member of Nvidia's Open Agent Safety Platform, which isolates agents in secure environments and can stop behavior through a hardware-level kill switch.
The strategy is both self-serving and surprising, given Microsoft's investments in OpenAI and Anthropic. But the logic is sound: monitors should be independent of the suppliers, just as auditors are independent of the organizations they audit. For Microsoft, that means turning its cloud security business into the enforcer of that separation.