News · AI Governance & Trust

Microsoft Extends Agent 365 to Rival Agent Platforms and Local Coding Agents

Agent 365 will issue Entra identities to agents on Salesforce, AWS, Google, Databricks, Snowflake and other platforms, and Defender can now find and block local coding agents. Much of it is in preview.

By The AI Enterprise Magazine Team5 min readNews · Weekly briefing

Illustration of a shield over a network of connected AI agents
Illustration of a shield over a network of connected AI agents

Key takeaways

  • Cross-platform identity. Agents on Salesforce, Amazon AgentCore, Google Vertex AI, Databricks, Snowflake, SAP, ServiceNow and others get Entra Agent IDs, with rollout finishing in December.
  • Local agents in scope. Defender discovery of local agents such as Claude Code is generally available; runtime protection is GA on Windows.
  • Autopilot arrives. A long-running agent with its own identity, memory and tools began rolling out to Frontier customers.
  • Check the status. Many governance features, including dynamic risk scores and multi-tenant management, remain in preview.

Microsoft used the opening day of Ignite 2026 in San Francisco on November 17 to push its agent control plane well beyond its own products. Agent 365, the registry and governance layer Microsoft launched in May, will now issue identities to agents running on rival platforms and will discover and police the local coding agents that developers install on their own machines. Alongside it, Microsoft began rolling out Autopilot, a long-running agent that lives inside a customer's tenant. For CIOs and security leaders, the practical question is whether one vendor's console can become the system of record for agents built everywhere else.

What did Microsoft add to Agent 365?

The biggest change is reach. According to Microsoft's Agent 365 blog, agents connected to Agent 365 receive a Microsoft Entra Agent ID, and that connection is rolling out across Salesforce Agentforce, Amazon AgentCore, Google Vertex AI, Databricks, Snowflake, Anthropic's Claude Managed Agents, OpenAI, Oracle, SAP, ServiceNow and UiPath, with the rollout due to finish in December. Microsoft says more than 50 million agents have been registered in Agent 365 since launch.

The blog lists a long set of supporting features, most of them in preview. A guided setup wizard and secretless authentication through Entra workload identity federation are meant to remove stored keys from agent connections, though Microsoft notes that capabilities vary by platform. A dynamic risk score gives each agent a single risk level that appears in the Microsoft 365 admin center, Defender, Entra and Purview. Multi-tenant management lets administrators compare, approve or block agents across tenants from one view. An export feature sends agent traces to Azure Monitor and Application Insights, with controls over which prompts and responses are included.

A tools registry is generally available, and Microsoft says connecting external catalogs such as GitHub and Unity Catalog will reach private preview in December. Plugin and skill governance is also generally available.

Why does local agent discovery matter?

Coding agents that run on laptops have been one of the harder blind spots for security teams, because they act with a developer's credentials and often sit outside any central registry. Microsoft says Defender and Entra Global Secure Access can now discover more than 60 local AI agents, naming Claude Code and GitHub Copilot CLI as examples. Local agent discovery in Defender is generally available, while network-level and shadow AI discovery remain in preview.

Defender runtime protection for local agents, covering Claude Code, GitHub Copilot CLI, OpenClaw and ChatGPT Desktop, is generally available on Windows with audit and block modes, according to the blog. Purview data loss prevention for those tools, Entra Agent ID for local agents and Intune controls for local agents and MCP servers are in preview. Microsoft also describes Conditional Access policies that are evaluated when an agent acts on a user's behalf, not only at sign-in, but it did not give a release status for that capability.

What is Autopilot?

Autopilot is a long-running agent for recurring work that Microsoft began rolling out to customers in its Frontier early-access program on November 17. The Copilot blog post by Jared Spataro says Autopilot "lives in your tenant and has its own identity, memory, and tools," and that each instance has its own Entra Agent ID and a named manager. The Agent 365 blog lists the extension of Entra Agent ID to Autopilot as a preview feature, so buyers should confirm which identity controls apply in their tenant before relying on them.

The same post said Copilot Home, which combines Chat and Cowork, and Office in Copilot are now generally available. Microsoft also said Copilot Code will reach general availability in the coming weeks and that Copilot Managed Runtime, which hosts apps built with Copilot, is generally available now.

How will companies measure what agents deliver?

Microsoft is adding a Business Analytics view to Agent 365 that it says helps leaders see where AI investments are delivering value. It is rolling out to Frontier customers. Analytics for Microsoft's own services require at least one Microsoft 365 Copilot license, while analytics for non-Microsoft agents need Agent 365 or Microsoft 365 E7 licensing. Both the Agent 365 blog and the Copilot post cite a BCG figure that more than 80% of AI spending sits outside the enterprise IT budget, which is the gap Microsoft says the feature is meant to close.

Microsoft did not publish dollar prices. The Copilot post describes a per-user subscription combined with usage-based consumption, and it cites a GigaOm report to argue that Copilot costs less per user than rival assistants. Those comparisons are vendor-supplied and should be treated as a starting point for negotiation rather than a benchmark.

Where else does Agent 365 show up?

The Ignite Book of News shows the governance layer being wired into infrastructure products as well. Azure SRE Agent gains Entra Agent ID and Agent 365 integration in preview, and the preview of Windows Server 2028 includes Agent 365 integration to help organizations govern AI. On the platform side, Microsoft said customers will be able to deploy open-weight models and agents from Foundry to their own Azure Kubernetes Service clusters, alongside an Agent Sandbox extension that runs agent-written code in VM-isolated sandboxes. Those AKS features are in preview.

What should enterprise leaders do now?

The direction is clear even though much of the release is in preview. Microsoft wants Entra to be the identity provider for agents regardless of where they run, in the same way it became the default directory for people. Organizations that already standardize on Entra will find that attractive, since it puts agent risk, data loss controls and access policies in tools their security teams already use.

Buyers with multi-cloud agent estates should test the connectors for the platforms they actually use, check which features are generally available versus preview, and decide where agent inventories and audit trails should live before a single vendor's registry becomes the default by inertia. Local coding agents deserve particular attention. If developers are already running them, discovery and runtime controls are worth piloting now, whichever vendor supplies them.

Frequently asked questions

What did Microsoft announce for Agent 365 at Ignite 2026?

Agent 365 now extends Entra Agent IDs to agents on platforms including Salesforce Agentforce, Amazon AgentCore, Google Vertex AI, Databricks and Snowflake, and adds discovery and controls for local coding agents. Many features are in preview.

What is Microsoft Autopilot?

A long-running agent for recurring work that lives in a customer tenant with its own identity, memory and tools. It began rolling out to Frontier program customers on November 17, 2026.

Can Microsoft Defender block local AI coding agents?

Yes. Defender runtime protection for local agents such as Claude Code, GitHub Copilot CLI and ChatGPT Desktop is generally available on Windows, with audit and block modes.

Filed under: AI Governance & Trust, Agentic AI, Security & Fraud

More from AI Enterprise Magazine

Read the magazine