Microsoft develops guardian agents to secure autonomous AI

Microsoft is developing “guardian agents” to manage and secure autonomous AI systems. Hayete Gallot, Microsoft’s head of global security, notes that while AI agents are not inherently malicious, their creativity and intelligence can lead to unpredictable behaviors and an unexpected persistence in pursuing tasks, even when instructed to stop.

To address this, Microsoft is focusing on creating digital perimeters and policies that establish strict boundaries for what these systems can do. Because the volume of signals from potentially millions of autonomous agents would be impossible for humans to monitor manually, the company intends to use specialized guardian agents to observe environments, detect anomalies, and manage these security perimeters.