Microsoft Treats AI Agents as Insider Risks With New Identity and Oversight Tools

Microsoft rolled out a cluster of new security features—Entra Agent ID, Dynamic Agent Risk scoring, Guardian Agents, and Project Perception—all built around the idea that autonomous AI agents inside a company's systems should be treated like employees who could pose an insider security risk, not just software.

Key points:

• Entra Agent ID gives each AI agent a verifiable digital identity, similar to how human employees get login credentials, so its actions can be tracked and attributed—like giving every AI agent its own employee badge instead of letting it borrow someone else's.

• Dynamic Agent Risk scoring continuously evaluates how risky an agent's behavior looks in real time.

• Guardian Agents are designed to supervise and intervene on other AI agents' actions.

This is a direct, practical response to the same underlying problem OpenAI admitted to this week and that the Cairn exposure illustrated: autonomous agents are hard to monitor with traditional security tools.

Microsoft is building AI-specific identity and oversight infrastructure before more customers deploy agents at scale—betting that treating agents as identities to be governed, rather than just software to be deployed, will become the enterprise standard.

Why It Matters: By treating AI agents as entities requiring identity management and behavioral monitoring, Microsoft is establishing what may become the enterprise standard for governing autonomous AI systems.