Microsoft expands Project Perception to monitor risky AI agents

Microsoft is expanding Project Perception, its security platform introduced in July, to monitor and contain AI agents with access to sensitive company data and systems, according to Redmond Magazine. The company calls these systems agentic insiders because they can use tools, access data and make decisions without constant human supervision.

Project Perception will discover agents inside company networks, assign them identities and permissions, monitor their activity across systems and continuously assess the risks from their actions. The platform entered public preview on August 3 and uses separate agents to find vulnerabilities, investigate suspicious activity and address security problems.

Microsoft also plans Guardian Agents that supervise other agents and intervene when their behavior becomes risky. These controls could restrict permissions or block specific actions before they are completed. Microsoft says future security tools will also need to explain why an AI model made a decision, rather than only record the resulting action.