Microsoft Introduces Guardian Agents to Watch Other AI Agents
Microsoft’s new security controls continuously assess enterprise agents and tighten their permissions when their behaviour becomes risky.
Microsoft has introduced Guardian Agents and adaptive security controls designed to monitor AI agents operating inside companies.
What happened
The new capabilities continuously assess what an agent is doing, what information it can reach and whether its risk level has changed. Permissions can be tightened dynamically according to the agent’s task, behaviour and company policy. Guardian Agents can also coordinate protective actions within boundaries set by security teams.
Why it matters
Enterprise agents can access data and take actions across multiple systems. If they receive excessive permissions or behave unexpectedly, they can create risks similar to an internal account compromise.
The bigger picture
AI-agent security is shifting from approving an agent once to monitoring it continuously. Identity, access control and automated containment are becoming core infrastructure for companies that want to deploy agents beyond simple chat interfaces.