Microsoft has introduced new security features aimed at protecting enterprises from risks posed by AI agents, headlined by Guardian Agents, a tool that monitors the agents a company deploys. It forms part of Project Perception, which Microsoft introduced in June with red, blue and green teams of agents to attack, detect and fix vulnerabilities from outside threats.
The new focus reflects warnings from CEO Satya Nadella, who wrote on his blog that enterprises should stop relying on AI model providers’ safety assurances and instead use independent systems to oversee agents, arguing that the supply of intelligence should be separated from authority over it. Security chief Hayete Gallot said frontier models cannot be trusted regardless of origin, not because they are malicious but because they are unpredictable, trained to complete tasks rather than follow organizational norms, and often granted broad access to sensitive data and tools.
Microsoft has also added agent detection to Microsoft Defender, identifying up to 60 agent types including Claude Code and Codex, alongside controls that limit agent access, adjust permissions by task and require human approval for risky actions. The stance is notable given Microsoft’s investments in both OpenAI and Anthropic. Microsoft is also an inaugural member of Nvidia’s Open Agent Safety Platform.
