Microsoft gives discovered local agents an Entra ID and a retirement clock

Microsoft is making agent identity automatic for the agents it finds on employee machines. At Ignite it announced that local agents — including Claude Code, OpenAI Codex and GitHub Copilot CLI — get their own Entra Agent ID when Defender and Global Secure Access discover them, and that lifecycle policies which disable an unsponsored or inactive agent are now generally available, with an out-of-the-box 60-day inactivity rule.

What happened

Microsoft’s Entra team published its Ignite announcements on 17 November 2026. The agent-related items fall into three groups.

Discovery and issuance. Microsoft Entra Agent ID for discovered local agents (preview) uses Microsoft Defender and Global Secure Access to find 60 local agent types, including Claude Code, OpenAI Codex and GitHub Copilot CLI. Each discovered agent receives an Agent ID so that its access can be governed and protected. Microsoft’s “digital teammate” agents get their own agent user account and identity, separate from the human they work for, so Conditional Access, Identity Protection and access governance apply to the agent’s own sign-ins.

Lifecycle and revocation. Automated lifecycle management of agent identities is now generally available. Policies check whether each agent has a human sponsor, is still needed and has been active recently; if not, Entra notifies the owner and later disables or deletes the agent according to the tenant’s instructions. The out-of-the-box policy disables agent identities that have been inactive for 60 days. Hard deletion protection (preview) can protect objects such as Agent IDs from permanent deletion, and Privileged Identity Management now extends just-in-time role activation to AI agents (preview).

Control at the network and MCP layer. Conditional Access policy plans, now in preview, ship templates for AI agents, and multitenant management covers Microsoft Agent 365 and Copilot agents (preview). Agent type-aware network controls for local agents (preview) apply different network policies per agent running on an employee device, and MCP policy can be enforced across servers, tools, resources, methods, protocol versions and connection security requirements. Microsoft also cited its own telemetry on Azure AI workloads: tool and agent abuse accounted for 12% of observed attack activity.

The framing in the post is blunt about the starting position: agents “inherit unnecessary privileges, persistent access, and weak governance”, and without central visibility agent sprawl becomes “shadow IT at a scale we’ve never had to manage before”.

Why it matters for KYA

The hard part of KYA has been issuance: an agent that nobody registered cannot be identified, scoped or revoked. Microsoft’s answer is to stop asking employees to register agents and instead discover them, then mint an identity. That converts an unmanaged population into directory objects with an owner field — which is exactly the substrate a KYA programme needs.

Two details deserve more attention than the feature list. The first is sponsorship. A lifecycle policy that asks “does this agent have a human sponsor” is a Know Your Agent control in the literal sense: it makes a named person accountable for the agent’s continued existence. Organisations that adopt the policy without deciding who may sponsor what will simply automate the answer “yes”.

The second is the 60-day inactivity default. Revocation is usually the weakest part of an agent programme — grants outlive the project that created them, and nobody notices until an audit. A default that retires idle agent identities is an unusually concrete control, and it will force teams to think about which agents are legitimately seasonal. A quarterly reporting agent that runs four times a year is not abandoned; it is idle by design, and it needs an exception or a longer window.

The limits are worth stating plainly. An Agent ID is a record in one tenant’s directory. It tells Microsoft’s control plane that the agent exists and who owns it; it does not tell a third-party API, a counterparty or a regulator that this agent is authorised by a particular principal for a particular task. Disabling an Agent ID is also not the same as revoking every credential the agent already obtained — that depends on token lifetimes and on whether downstream services check back. Discovery is a prerequisite for KYA, not a substitute for it.

For practitioners, the practical sequence is: turn on discovery and get the inventory; require a sponsor for every agent identity, including the ones discovered on laptops; set a lifecycle policy that matches how your agents actually run; export the inventory with sponsors and last-activity dates as your agent register; and extend the same register to agents outside Microsoft’s estate, because the ones Microsoft cannot see are the ones that will surprise you.

What to watch

  • Whether an Agent ID carries attributes a third party can verify — sponsor, publisher, permitted scope — or remains a tenant-local object that only Microsoft’s own services understand.
  • Whether the 60-day default survives production. Expect a wave of exceptions for seasonal and on-call agents, and watch whether Microsoft makes the window configurable per agent class.
  • Whether MCP policy enforcement covers non-Microsoft MCP clients and servers, or only the ones Microsoft ships or brokers.
  • Whether “digital teammate” identities become portable across clouds. An agent that acts for a person in two ecosystems needs one accountable identity, not two unrelated directory entries.

Sources

  • Microsoft Tech Community, “Microsoft Entra news announcements from Microsoft Ignite November 2026”, 17 November 2026 — techcommunity.microsoft.com
  • Microsoft Learn, Microsoft Entra Agent ID documentation — learn.microsoft.com
  • Microsoft Ignite 2026, session BRK343 (Entra announcements) and BRK339 (local agents) — ignite.microsoft.com

Know Your Agent (KYA) explains agent identity, verification and accountability. This is an explainer, not legal or compliance advice — see our Sources & methodology. New to KYA? Start with What is Know Your Agent? and the glossary.