Microsoft adds task based controls to govern AI agent access

Microsoft’s November 2026 update outlines a wave of new protections and management capabilities across Defender, Intune and AI agent governance. The core change is Dynamic Permissions, which gives each AI agent task-scoped authority and adjusts access in real time as risk signals evolve.

Microsoft explains Dynamic Agent Risk will assess whether an agent’s actions line up with its current task, while Dynamic Conditional Access can tighten permissions, restrict resources, or prompt human approval for higher-impact steps. The company frames this as a way to keep agent authority aligned with task and risk throughout execution.

Microsoft Execution Containers (MXC) are now generally available, providing an OS-enforced containment layer to limit what AI agents can access, with least-privilege controls across files, networks, processes and user interactions. Integration with Intune, Entra and Microsoft Agent 365 is planned to extend governance and attribution enterprise-wide.

In parallel, Microsoft Intune Endpoint Privilege Management gains new system settings support so standard users can alter IPv4, IPv6 and DNS server values on their devices, with administrators deciding how those changes are elevated and whether a business justification or Windows authentication is required. Defender Experts MDR gains a consolidated overview in the Defender portal, enabling teams to review hunts and Defender Experts findings more efficiently.

Other updates include the enhanced Defender for SQL Servers agent in Azure Government for SQL protection, discovery of locally running AI agents and their MCPs on Windows and macOS, and a Security Exposure Management focus via Secure Now to prioritise foundational controls. Microsoft Ignite is slated for 17–20 November 2026 in San Francisco or online.