Microsoft is extending its Project Perception security framework to monitor AI agents inside businesses, seeking to limit risks when software can access company data, use tools and take action across corporate systems.
The company outlined the approach Oct. 19. Microsoft Security Executive Vice President Hayete Gallot said agents with broad access can stray from an organization’s intended task even without being compromised by an attacker. That makes an agent’s behavior after it receives access a concern for security teams, she said.
Project Perception is intended to help teams find agents and reassess their risk while they work. Microsoft said the system would use security telemetry to identify agents and update its assessment of their risk during operation. The company introduced Project Perception in July as part of its work against AI-driven attacks and is now extending it to agents used within businesses.
Microsoft said Guardian Agents would supervise other agents’ activity and carry out ongoing security work. Separate permissions and access controls could change what an agent is allowed to do as risk changes. For example, a company could let an agent read data while blocking changes to it, require a person to approve a consequential action or stop the agent.
Nadim Abdo, Microsoft’s corporate vice president for identity and network access, said people would set the policies and approval boundaries governing that supervision. Microsoft also said it is expanding discovery to 35 additional types of locally running agents, bringing the total to 60 supported local agents. They include Claude Code, OpenAI Codex and GitHub Copilot.
Gallot called for greater visibility into an agent’s reasoning so security teams can judge whether its actions still fit the assigned task. Microsoft did not give a date for that step.
