Microsoft Announces New Controls to Secure AI Agents
Summary
Microsoft announced agent discovery, identity, task-scoped permissions, dynamic risk controls, and Guardian Agents to help organizations monitor and contain AI agents as their behavior and risk change.
Key points
- Microsoft says its expanded discovery supports 60 types of local agents, including Claude Code, OpenAI Codex, and GitHub Copilot.
- Microsoft is extending Entra Agent ID to cover additional agent patterns, including local and previously unknown agents discovered by security sensors.
- Its announced Dynamic Permissions model gives agents task-scoped, time-bound access instead of automatically inheriting their users’ standing permissions.
- Dynamic Agent Risk and Dynamic Conditional Access are intended to adjust access as risk changes, from restricting writes or requiring human approval to isolating or stopping an agent.
- Microsoft says controlled runtimes, policy-enforced tool and resource access, and independent security controls should limit an agent’s ability to bypass safeguards.
- Project Perception’s announced Guardian Agents and Playbooks are intended to investigate risk changes and coordinate protective actions within boundaries set by people.
Article Details
- Event Type
- Microsoft announced security capabilities and product updates for discovering, identifying, containing, and managing risk from AI agents, including expanded Project Perception, Guardian Agents, Dynamic Permissions, Dynamic Agent Risk, and Dynamic Conditional Access.
- Impact
- The announced capabilities are intended to help organizations monitor agent activity, limit agents to task-scoped permissions, and adapt or contain access when risk changes. The article describes these as product announcements and intended protections; it does not report a specific security incident or confirmed customer impact.
People
Satya NadellaMicrosoft Chairman and Chief Executive Officer; cited for describing the trust architecture principle of separating intelligence from authority.Terrell CoxMicrosoft Corporate Vice President and Deputy Chief Information Security Officer, Customer Security; described Microsoft’s experience securing agents.
Vendors
Products
Microsoft Agent 365Project Perception and Microsoft Agent 365 share signals, context, and a unified registry giving IT and security teams a consistent view of agents and risk, with the controls each needs.Microsoft Entra Agent IDWe are extending Microsoft Entra Agent ID with identity types for additional agent patterns, including local and previously unknown agents discovered through security sensors.Microsoft Entra Conditional AccessDynamic Conditional Access extends Microsoft Entra Conditional Access for agents so a change in risk can drive coordinated action across multiple enforcement points.Microsoft Execution Container (MXC)Sandboxes such as Microsoft Execution Container (MXC) provide policy-controlled access to resources the agent needs to complete its task.Project PerceptionProject Perception—Securing your agentic insider
Tools
Claude CodeWe are expanding discovery to 35 additional types of local agents, bringing the total to 60 supported local agents, including Claude Code, OpenAI Codex, and GitHub Copilot.GitHub CopilotWe are expanding discovery to 35 additional types of local agents, bringing the total to 60 supported local agents, including Claude Code, OpenAI Codex, and GitHub Copilot.OpenAI CodexWe are expanding discovery to 35 additional types of local agents, bringing the total to 60 supported local agents, including Claude Code, OpenAI Codex, and GitHub Copilot.