Microsoft announces Guardian Agents to constrain AI agent permissions
Original title: Microsoft wants to limit AI agents’ permissions when they stray from the job
Discovery now covers 60 local agent types, and risk signals can revoke access or stop an agent.
Microsoft announced Guardian Agents and Playbooks as an extension of Project Perception, tying agent oversight to Agent 365 and Entra Agent ID. The design gives agents identities separate from users, task-scoped Dynamic Permissions that cannot exceed the user's rights, and Dynamic Agent Risk that can limit writes, shorten access, require human approval, isolate a runtime or stop execution. Discovery expands by 35 local agent types to 60, including Claude Code, OpenAI Codex and GitHub Copilot. People keep control of objectives and consequential decisions.
- Dynamic Permissions last only for an approved task and follow organization-wide policy.
- Higher-impact actions can require verified human approval.
- Conditional access can keep read rights while blocking writes as risk rises.
- An example expense agent could lose approval rights after following suspicious instructions.
- Vendors
- Microsoft
From the original article
4 min read at helpnetsecurity.com
Microsoft’s Guardian Agents in Project Perception adapt as agent activity, risk, security context and organizational policy change. They help security teams understand why risk changed, investigate relevant evidence, recommend or apply containment measures and use what they learn to strengthen policies. People retain control over objectives and consequential decisions.
By Anamarija Pogorelec · helpnetsecurity.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.