Microsoft has introduced new security features to protect enterprises from risks associated with AI agents. The company launched Guardian Agents, a monitoring tool that is part of the Project Perception suite, which was originally introduced in June.
Project Perception includes three classes of agents: red, blue, and green teams designed to simulate attacks, detect vulnerabilities, and implement fixes. These tools were initially created to defend against external hacking models.
CEO Satya Nadella now warns that enterprises must stop trusting AI model providers. He advocates for separating the supply of intelligence from the authority over it, suggesting that companies should use independent systems to monitor their deployed agents.
Microsoft Defender now includes a detection feature capable of identifying 60 types of agents, such as Claude Code, Codex, and OpenClaw. The software can restrict access to data and require human approval for risky actions.
The company is also a member of Nvidia’s Open Agent Safety Platform, which uses hardware to isolate and monitor agents. Additionally, Microsoft has launched Surface Laptop Ultra PCs that feature built-in security controls for agent management.
