Microsoft adds controls for third-party and local AI agents in Agent 365
The Ignite update connects more outside platforms and discovers agents on employee devices. Availability varies, and advanced third-party management still requires a premium license.
At Ignite on November 17, 2026, Microsoft expanded Agent 365 into a management layer for supported third-party platforms and employee-run local agents. It can give connected agents Microsoft Entra Agent IDs, inventory their owners and tools, and apply supported access and risk controls; discovery can surface more than 60 local agents. Coverage is uneven: many cross-platform and shadow-AI controls remain in preview, integrations are rolling out through December, and capabilities vary by platform.
01
Microsoft says organizations have registered more than 50 million agents in Agent 365 since its May 2026 launch.
02
Connected-platform support includes Salesforce Agentforce, Amazon AgentCore, Google Vertex AI, Databricks, Snowflake, Anthropic Claude Managed Agents and OpenAI, with rollout completion due in December.
03
Defender discovery is generally available on managed Windows and macOS devices, while network-level discovery and registry entries for unapproved agents remain in preview.
Microsoft is bringing more outside AI agents—and agents running on employee computers—into its corporate management systems. Its November 17, 2026, Agent 365 announcement at Ignite adds broader identity support, local discovery and cross-organization controls. Some capabilities are generally available; others remain in preview or have rollout milestones scheduled for December.
An identity beyond Microsoft’s own platforms
Agent 365’s connected-platform feature brings agents from supported services into a shared registry without requiring changes to their code. The inventory records what an agent is, where it was created, who owns it and which tools it uses. Activity details, including model and tool calls, are available on supported platforms.
The update gives connected agents a Microsoft Entra Agent ID, distinguishing them from people in the identity inventory. Organizations can use that identity to manage an agent’s lifecycle, apply supported access and security policies, assess identity risk and revoke access.
Microsoft says support is rolling out across platforms including Salesforce Agentforce, Amazon AgentCore, Google Vertex AI, Databricks, Snowflake, Anthropic Claude Managed Agents and OpenAI, with completion due in December. The connected-platform updates are in preview, and capabilities vary by platform.
A guided setup wizard walks administrators through connections. For supported platforms, secretless authentication removes the need to create, store and rotate long-lived credentials for those connections.
Agent registrations since launch
More than 50 millionAgents registered
Microsoft says organizations have registered more than 50 million agents in Agent 365 since its May 2026 launch.
Finding agents employees bring themselves
Microsoft says Defender and Entra Global Secure Access together can discover more than 60 local AI agents, including Claude Code and GitHub Copilot CLI, and their configured MCP servers. Defender’s discovery is generally available on managed Windows and macOS devices, linking each agent to its device and user.
Global Secure Access adds network visibility into agents’ destinations, servers and resources. Agent 365 uses discovery signals to add local and unapproved agents to its registry. Both network-level discovery and discovery of this unapproved activity—often called shadow AI—in the registry remain in preview.
A separate preview assigns identities to supported discovered local agents, starting with GitHub Copilot CLI and Claude Code. Microsoft says agent-specific controls can restrict a risky agent while preserving the employee’s own access. A continuously updated risk view across Microsoft’s administration, security, identity and compliance tools is also in preview.
From inventory to operating decisions
Multi-tenant management, in preview, gives partners and enterprises a shared view across the separate Microsoft environments they administer. Administrators can compare an agent’s configuration and access between environments, review risk details, and approve or block agents at scale.
Tools registry: A centralized inventory of agents’ tools and connections is generally available. Connections to external catalogs are planned for private preview in December.
Activity export: Administrators can continuously send detailed agent activity records to Azure Monitor and Application Insights, filtering by agent or platform and choosing which prompts and responses to include.
Usage and value: Agent Map’s preview classifies sessions by the work performed to help identify overlapping agents. Business Analytics offers usage, spending forecasts and recommendations in Frontier preview.
The licensing change draws a boundary around that broader reach. Qualifying Microsoft 365 subscriptions now include core usage and activity insights for Microsoft agents, including active users, sessions, exception rates and runtime. Premium capabilities—including non-Microsoft agent management insights and advanced governance and security—still require an Agent 365 or ME7 license.
Reader comments
Newest comments first. Replies stay oldest first.