Microsoft has introduced new security features to help enterprises manage risks associated with AI agents, including the Guardian Agents tool. This feature is part of Project Perception, launched in June to monitor AI behavior and address potential vulnerabilities.
The initiative responds to concerns from CEO Satya Nadella regarding 'insider threats' posed by autonomous AI agents. Nadella advises companies to maintain independent oversight systems separate from AI model providers, even when those providers offer their own security guarantees.
Microsoft's security head, Hayete Gallot, noted that frontier models are inherently unpredictable and may prioritize task completion over organizational norms. These agents often possess high-level access to sensitive data and critical business tools.
The company has also updated Microsoft Defender to detect up to 60 types of AI agents, such as Claude Code and Codex. New controls allow administrators to restrict agent permissions and mandate human approval for high-risk operations.
Microsoft is a founding member of Nvidia's Open Agent Safety Platform, which currently has support from over 100 technology firms. The platform uses hardware-based monitoring to isolate agents and terminate abnormal activities.
