Enterprise AI
Microsoft Ignite 2026: agents get identities, owners and audit trails
The announcements that matter were not about what agents can do. They were about who they are, what they are allowed to touch, and who answers when they get it wrong.
In brief
At Ignite 2026 (17–20 November) Microsoft's agent announcements centred on governance rather than capability: Agent 365, Entra Agent IDs extended to locally running agents, Purview DLP inspection before agents process data, and Copilot Managed Runtime. Availability is split across GA, preview and December private preview.
What happened
Microsoft used Ignite 2026 to describe an enterprise agent estate and the controls around it. Plugin and skill governance for Copilot and Cowork reached general availability, with governance for MCP servers and for shadow and local agents in preview. Agent 365 activity insights went GA, while Dynamic Agent Risk and Cost Management entered preview. An action centre for inactive and ownerless agents is expected in private preview in December.
On identity, Microsoft Entra began assigning Agent IDs to discovered local agents, starting with GitHub Copilot CLI and Claude Code, extending an approach already applied to Copilot Studio and Foundry agents. Microsoft Defender extended runtime protection to local agents, and Microsoft Purview data loss prevention gained the ability to detect and block sensitive information before agents process, expose or transmit it. Intune is previewing governance for local AI agents on managed devices using Microsoft Execution Containers.
Copilot Managed Runtime reached general availability, and Microsoft IQ — described as turning organisational context into an asset agents can reuse — also went GA. Status labels differ across secondary sources and some items were still moving during the event; verify against Microsoft's own Book of News before planning around a specific date.
Why it matters
An agent that can act inside your systems is not a feature. It is a principal — something that holds permissions, takes actions and leaves consequences. Enterprises already have decades of machinery for managing principals: identity, least privilege, audit, joiners-movers-leavers. What they have not had is any of it pointed at software that improvises.
The telling detail is Entra issuing identities to agents Microsoft did not ship, including a competitor's. That is an admission that the agents running inside enterprises will not come from one vendor, and that governance has to start from discovery rather than from a catalogue.
Business and industry implications
Shadow agents are the near-term exposure. Staff are already running coding and research agents with their own credentials, on managed devices, against real data. Most organisations cannot currently enumerate them. Discovery and identity assignment are unglamorous, and they are the prerequisite for everything else.
The DLP change is the one with immediate operational weight: inspecting content before an agent processes it, rather than after it has acted, is the difference between a control and a report.
Leadership perspective
Give every agent an owner, a scope and a measurable outcome before you give it autonomy. The governance question is not philosophical — it is the ordinary question of who is accountable, asked about a new kind of actor. Organisations that answer it now will be able to expand agent use deliberately. Those that do not will discover their agent estate through an incident.
Three questions for business leaders
- Can we enumerate every agent running against our data today, including ones we did not deploy?
- Who is the named owner of each agent's actions, and how are exceptions escalated to a human?
- Are we measuring completed business outcomes, or only usage?