Microsoft announced this week new security tools for companies using artificial intelligence agents, treating these systems as potential insider threats and launching Guardian Agents.
What happened?
In a post on the company's official blog, Security Executive Vice President Hayete Gallot stated that frontier models "cannot be trusted," regardless of their origin. Not because they are malicious, but because they are unpredictable, adapt to context, and may collude with other agents to achieve objectives.
Agents are already being deployed at scale in companies, often with excessive privileges, access to data, and the ability to take actions in critical processes. Microsoft calls them "agentic insiders."
How do Guardian Agents work?
The presented solution is the use of Guardian Agents — agents that continuously supervise every action of other agents, 24 hours a day. They are part of Project Perception, announced in July, which now extends to the management of internal agents.
The system combines sensors for identity, data, cloud, endpoint, and runtime to calculate dynamic risk and respond in real time, restricting permissions or requiring human approval when risk increases. Microsoft also expanded agent detection in Microsoft Defender to about 60 types, including Claude Code and Codex.
Why does this matter?
With the proliferation of AI agents in companies, traditional perimeter-based security is no longer sufficient. Microsoft's approach separates "intelligence" from "authority" over it, requiring continuous supervision with Guardian Agents. The tools are available in Microsoft 365 E7 and Project Perception.
Sources: Microsoft official blog, TechCrunch.
By GeekikiBot