Organizations have registered more than 50 million agents in Microsoft Agent 365 since its May 2026 launch. That scale brings new questions: who owns each agent, what risk do they carry, and what do they cost – and importantly, is it worth it. Today at Microsoft Ignite 2026, we're expanding Agent 365, the observability platform for AI, to give you more visibility and control across your AI estate. IT, security, and business leaders now work from shared context on Microsoft and non-Microsoft agents, so they can scale AI together, each in the tools they already use. IT keeps ownership, tools, and lifecycle clear, security contains agent risk, and business leaders see what AI costs and what it returns.
There's a lot to explore in Agent 365 at Ignite, and Microsoft’s broader approach to securing AI. Here are five highlights you won't want to miss:
- Agent 365 Business Analytics: see where AI investments are delivering value across the org and for your team, with usage, spend forecasting and recommendations (Frontier preview).
- Connected platforms, now with Microsoft Entra Agent ID for more partners: in addition to Amazon Bedrock agents, agents from Google, Salesforce, Databricks, Snowflake, Anthropic, OpenAI, and more now get their own identity in Microsoft Entra, so IT and security can govern them alongside Microsoft agents. A guided connection wizard and secretless authentication make connecting faster and remove stored keys (Preview; capabilities vary by platform).
- Dynamic agent risk moves to preview: one continuously updated view of agent risk, visible in the Microsoft 365 admin center, Microsoft Defender, Microsoft Entra, and Microsoft Purview (Preview).
- Multi-tenant agent management: partners and enterprises manage agents across all their tenants from one place, deciding which agents run where and who can use them. (Public Preview)
- Local agents, secured end to end: New capabilities across Microsoft Agent 365, Microsoft Security, and Windows help organizations discover local agents, contain their execution, govern access, protect sensitive data, and investigate and recover from threats. Agent-specific identity and risk controls help security teams restrict a risky agent while preserving the employee’s own access. Learn more about specific announcements below.
Go directly to the announcements that matter most to you and your organization:
- Management: Manage your end-to-end AI estate from one platform
- Governance: Operate with clear ownership, permissions, and accountability
- Security: Understand AI risk, enforce controls, and keep agents protected
- Business Value: Optimize AI spend and turn your investment into business value
Management: Manage your end-to-end AI estate from one platform
You can't scale AI without trust, and you can’t trust what can't see. These updates bring agents from Microsoft platforms, partner platforms, and employee devices, plus the tools they use, into one view in Agent 365, so IT, security, and business leaders work from the same picture of what's running and how it's used.
Connect your agent estate across platforms
Many organizations build agents on more than one platform. A sales team might build in Salesforce Agentforce, a data team in Databricks Genie, and developers in Amazon Bedrock. Each platform may have its own console and its own record of its agents, so IT and security have to work in each one to see what's running and take actions to manage and control it.
With connected platforms, you connect a supported platform once, and Agent 365 syncs its agents into the registry alongside your Microsoft agents, with no changes to agent code. Connected agents bring important insights, including what the agent is, where it was created, who owns it, and the tools it uses. On support platforms you can also see the agent’s activity, what the agent was asked, what it did, which models, tools, or agents it called, and whether it succeeded. These insights give IT, security, and business leaders the context and evidence to manage, govern, secure, and get value from these agents, not just see them. At Ignite, three new updates build on that foundation:
- Entra Agent ID for more platforms: Identity is the foundation for governance and security. Connected agents now automatically receive a Microsoft Entra Agent ID, which brings them under identity management and helps IT and security tell agents apart from people in a shared identity inventory. With that identity, organizations can govern each agent's lifecycle, apply supported Conditional Access and security policies, assess identity risk, and revoke access when needed. Support is rolling out across Salesforce Agentforce, Amazon AgentCore, Google Vertex AI, Databricks, Snowflake, Anthropic Claude Managed Agents, OpenAI, Oracle, SAP, ServiceNow, and UiPath, and will complete in December.
- Guided setup for platform connections: Connecting a platform means setting things up in both Agent 365 and the source platform: choosing what to connect, granting access, and verifying the connection. A new step-by-step wizard walks admins through connecting a platform, tools registry, or gateway to Agent 365 so setup is easier to follow from start to finish.
- No stored secrets for supported connections: Connections that rely on stored credentials need someone to create, store, and rotate them, and each one is another credential to protect. Secretless authentication uses Microsoft Entra workload identity federation to connect Agent 365 to supported platforms, removing the need to create, store, and rotate long-lived secrets for those connections.
Learn more about connected platforms.
Agent 365 now connects to more platforms, including Salesforce Agentforce, Amazon AgentCore, Google Vertex AI, Databricks, and Snowflake, with each agent getting a Microsoft Entra Agent ID.
Local and shadow agent discovery
Some of the agents in your organization never went through IT. Employees run agents locally on their own devices or bring them in without approval, so managing your AI estate also means managing the agents you may not even know about. Together, Microsoft Defender and Microsoft Entra Global Secure Access now discover more than 60 local AI agents, such as Claude Code and GitHub Copilot CLI, and their configured MCP servers. Defender supports managed Windows and macOS devices and ties each agent to its device and user, while Global Secure Access adds network-level visibility into each agent’s destinations, MCP servers, and resources. Agent 365 uses these signals to bring the local and shadow agents into the Agent registry. IT and security teams see AI running outside approved channels, often called shadow AI, in the same inventory they use for other agents. Local agent discovery in Microsoft Defender is now generally available, and network-level discovery in Microsoft Entra Global Secure Access and shadow AI discovery in the Agent registry are in preview.
Learn more about discovery of local agents and shadow agents in Agent 365.
Local and shadow agents, such as Claude Code and GitHub Copilot CLI, discovered by Defender and Global Secure Access and added to the Agent 365 registry.
Secure local agents with Entra Agent ID
Microsoft Entra Agent ID automatically assigns an identity to supported local agents discovered by Microsoft Defender on managed devices and by Microsoft Entra Global Secure Access on the network, starting with GitHub Copilot CLI and Claude Code. With a distinct identity for each local agent, organizations can tell agent activity apart from human activity and apply supported access controls. These capabilities are in preview.
Learn more about Microsoft Entra Agent ID
Microsoft Entra Agent ID gives each discovered local agent, starting with GitHub Copilot CLI and Claude Code, its own identity, separate from the user's.
Extend Entra Agent ID to Microsoft Autopilot
As organizations adopt more cloud agents, admins need a consistent way to manage agent identity and access across services. Microsoft Entra Agent ID now extends to Microsoft Autopilot, building on existing support for agents built with Copilot Studio and Microsoft Foundry. This gives admins a consistent way to identify, authenticate, and govern these agents. This capability is in preview.
Learn more about Microsoft Entra Agent ID
Export Agent Activity
Troubleshooting agent behavior requires detailed traces in the monitoring tools teams already use, with control over which prompts and responses are shared. The new Export Agent Activity feature allows admins to continuously sync agent traces from Agent 365 into their own Azure Monitor and Application Insights. They can filter by agent or platform and decide which prompts and responses are included. Developers, builders and SRE teams can also debug agent runs, model calls and tool calls with their existing dashboards and queries.
Learn more about exporting agent activity
Export Agent Activity syncs agent traces from Agent 365 into Azure Monitor and Application Insights, with filters for agent or platform and control over which prompts and responses are included.
Tools Registry
Running AI at scale takes observability into more than the finished agents themselves. You also need to see the pieces their behavior builds on: the tools they use and the resources and services they connect to. The Agent 365 tools registry gives IT one centralized view of those pieces, so they can see which tools agents use and spot commonly used or potentially risky tools and connections. IT can also connect external catalogs, such as GitHub, Unity Catalog (Databricks), and gateways, so Microsoft and non-Microsoft tools appear in the same view. The tools registry is generally available today, and connecting external catalogs is coming to private preview in December.
Learn more about Tools Registry
The Agent 365 tools registry gives IT one view of the tools agents use, including commonly used and potentially risky tools and connections.
Agent Map Use Cases
As the number of agents grows, it becomes increasingly important to understand what work they are doing. That knowledge helps you avoid duplicated agents and focus your investment where it pays off. Agent map use cases classify agent sessions by the type of work performed, helping admins understand how agents are being used across the organization. These insights make it easier to identify adoption patterns, uncover overlapping or duplicative agents, and understand where agents are delivering the most value, helping organizations optimize investments as their agent estate grows. These capabilities are now in preview.
Learn more about the Agent Map
Agent map use cases classify agent sessions by the type of work performed, helping admins spot adoption patterns and overlapping agents.
Manage agents across multiple tenants at scale
Partners manage many customer tenants, and some enterprises run separate tenants for their subsidiaries, regions, or acquisitions. Administering and governing agents one tenant at a time doesn't scale. Multi-tenant agent management provides partners and multi-tenant organizations with one view of Microsoft and non-Microsoft agents across the tenants they manage in Agent 365, within the Microsoft 365 admin center. From this centralized experience, admins can deploy agents to multiple tenants easily, compare how the same agent is configured across tenants, and who can access it in each tenant, then review its security, compliance, and risk details wherever it's deployed, and approve or block agents at scale. Partners can run agent management as a repeatable service across customers, allowing central IT to make each agent decision once instead of once per tenant. Multi-tenant agent management is in preview.
Learn more in the multi-tenant management blog
Multi-tenant agent management gives partners and multi-tenant organizations one view of agents across their tenants, to compare configuration and access and approve or block agents at scale.
Expand usage insights for Microsoft agents to Agent 365 core users
Microsoft is expanding access to Agent 365’s usage and activity insights in the Microsoft 365 admin center for Microsoft pre-built agents and agents built on Microsoft platforms. This change will provide additional core Agent 365 capabilities, including the agent overview, active users, and advanced usage metrics for activity, sessions, exception rate, run time, to users with qualifying Microsoft 365 subscriptions. This gives admins foundational visibility into agent usage and activity, and helps them understand where additional management, governance, and security capabilities may be needed. This licensing change is now GA.
To unlock premium capabilities including non-Microsoft agent management insights, and advanced governance and security capabilities, customers will still require an Agent 365 or ME7 license.
Learn more about what’s included in Microsoft 365 subscriptions. (Microsoft Agent 365 - Service Descriptions | Microsoft Learn)
Agent overview, active users, and usage metrics for Microsoft agents in the Microsoft 365 admin center, now available with qualifying Microsoft 365 subscriptions.
Governance: Operate with clear ownership, permissions, and accountability
Agents and their tools multiply faster than manual reviews can keep up. These updates help organizations say yes to agents with confidence, with clear rules for what agents can use and do.
Plugins, skills, and MCP server governance
The tools AI can reach decide what it can do. Administrators decide which plugins and skills are available in Microsoft 365 Copilot and Copilot Cowork, and assign them to specific users and groups. Expanded governance now brings MCP servers from Azure AI Gateway and third-party gateways, such as LiteLLM, into the same experience. Before blocking an MCP server, admins see which agents use it and which plugins depend on it, so they can block it without breaking the plugins people rely on. Plugins and skills governance is generally available, and Azure AI Gateway and LiteLLM gateway support is in preview.
See plugins, skills, MCP servers, and connectors in one place in the Microsoft 365 Admin Center tool registry.For each MCP server, see usage, the agents using it, exceptions, and calls over time.
Learn more about tools governance.
Local and shadow agent governance
Administrators can review discovered local agents and their MCP connections to decide which agents and tools to allow, restrict, or investigate. They can block unsanctioned local agents from the Agent registry using controls in Microsoft Intune, Microsoft Defender, and Microsoft Entra Global Secure Access. For approved agents, Microsoft Intune policies and Microsoft Execution Containers (MXC) on Windows help constrain tool execution, while Microsoft Entra provides task-scoped permissions and risk-based access controls. Local and shadow agent governance is now in preview.
Copilot Cowork with Agent 365 plugin
Managing your AI estate can now start with a plain-language question or request. Say what you want to know or do, and Copilot Cowork takes it from there. With the Agent 365 plugin for Cowork, administrators inspect their AI estate and automate governance by describing the outcome they want. Examples include keeping a departing employee's agents running or containing agents exposed to a compromised connector. Cowork pauses for admin approval before it takes consequential actions, so admins save manual steps and stay in control of each change. Copilot Cowork with Agent 365 is now in preview.
Ask Copilot Cowork about your agents, and the Agent 365 plugin returns an inventory you can act on, including blocking an agent.
Action center
As agents multiply, a big part of governance is knowing where to start. An agent without an owner has no one accountable for it, and an inactive agent can keep access it no longer needs. Action center brings governance tasks into one focused experience, so administrators start with the agents that need attention rather than searching the full registry. It begins by surfacing inactive and ownerless agents for review and cleanup. Action center will be available in preview in December.
Security: Understand AI risk, enforce controls, and keep agents protected
Agents act with access to data, tools, and systems, which makes them a new kind of insider. Securing them is continuous work: security teams need to know which agents exist, contain what each agent can do, protect what's in its path, and adapt protection as risk changes. Agent 365 and Microsoft Perception share signals, context, and a unified registry, so IT and security work from the same view.
Read how Microsoft Security approaches securing the agentic insider.
Dynamic agent risk
Dynamic agent risk combines runtime signals across identity, access, data, endpoints, cloud, and network with exposure and blast radius to deliver one risk level per agent. The assessment updates as the agent takes new actions, becomes a target, or receives code or model updates. It appears in the Agent 365 registry in the Microsoft 365 admin center, and the same risk level shows as well as in Microsoft Defender, Microsoft Entra, and Microsoft Purview, so security teams can act on riskiest agents first. Microsoft is also working with partners such as Check Point, Cyera, Darktrace, and Zenity to add their runtime signals through Microsoft Sentinel. Dynamic agent risk is now in preview, and in expanded limited preview within Microsoft Perception.
Identity and access controls that adapt to risk
Microsoft Entra Conditional Access dynamic controls evaluate policy at the point of action, not only at sign-in, so as agent risk and context change, organizations can restrict sensitive operations or require human approval while allowing lower-risk work to continue. Conditional Access also now lets admins set access policies for agents acting on a user's behalf, separate from that user's own policies. Together, these new capabilities help organizations apply more precise, adaptive protections to supported agent scenarios.
Learn more about Conditional Access for agents
Data security for Microsoft and local agents
Microsoft Purview Data loss prevention can now detect and block sensitive information before Microsoft agents and key local agents process, expose, or transmit it. Key local agents include Claude Code, Claude Desktop, GitHub Copilot CLI, the GitHub Copilot app, and the Visual Studio Code extensions for GitHub Copilot and Claude Code. Flexible policy scoping lets administrators apply security policies to individual agents, groups of agents, or the whole organization. Data protection follows agents in the cloud and on supported devices. Data loss prevention for Microsoft agents and key local agents, and flexible policy scoping, are now in preview in the Microsoft Purview portal.
Learn more about Microsoft Purview for AI
Runtime protection for local agents
Microsoft Defender extends runtime protection for local agents. Now generally available on Windows, Defender inspects critical points in the agent loop and can block malicious instructions before the agent acts. Supported agents include Claude Code, GitHub Copilot CLI, OpenClaw, ChatGPT Desktop (Chat) and more.
Each detection raises an alert with agent and session context, correlated into incidents for your SOC; in Block mode, Defender also stops the action and notifies the user. Configure Audit or Block mode in the Microsoft Defender portal or Microsoft Intune. With Network DLP, Microsoft Entra and Microsoft Purview help keep sensitive data from leaving over the network.
Security posture for third-party agents and tools
Microsoft Defender now extends extends security posture assessment to agents built on Amazon Bedrock, Google Gemini, Google Vertex AI, and Anthropic Claude. Now in preview, they appear in the same AI agent inventory as your Microsoft Copilot Studio and Microsoft Foundry agents, with risk assessments and security recommendations, so you can prioritize risk across platforms. Bring them into scope by connecting their platforms to Agent 365 through connected platforms.
Learn more about agent security posture in Defender.
Endpoint and network controls for local agents
Microsoft Intune policies define the file and network boundaries that Microsoft Execution Containers (MXC) on Windows enforce for supported tool execution. Those boundaries are enforced outside the agent’s control, so the agent cannot grant itself more access. Device-wide restrictions remain a separate option: Intune can mark devices with unapproved agents as noncompliant so Microsoft Entra Conditional Access can block those devices from company resources.
Microsoft Entra Global Secure Access adds network visibility and control over local agents, cloud agents, and MCP servers, including MCP Firewall and network policies tailored to agent type and risk. IT and security act on agent risk at the device and the network as well as in the cloud. Intune controls for local agents and Global Secure Access controls for agents are now in preview.
Learn more about Microsoft Intune
Intune and Global Secure Access controls set file and network boundaries for local agents and MCP servers, enforced outside the agent's control.
Business Value: Optimize AI spend and turn your investment into business value
AI costs can grow faster than budgets, and usage alone doesn't show whether an agent earns its keep. Leaders need to plan, control, and optimize AI spend, and connect that spend to outcomes. With over 80% of AI spend now sitting outside the enterprise IT budget, it’s critical that IT and business leaders work together to understand AI usage, costs and value to maximize business impact.
Agent 365 Business Analytics
Agent 365 Business Analytics helps business leaders see where AI investments are delivering value, across Microsoft and non-Microsoft AI costs and outcomes. Built on the Agent 365 observability foundation, Business Analytics brings AI spend, usage, and value together in one place, helping leaders understand their spend, what value it's delivering, and where to invest next.
Agent 365 Business Analytics overview page,bringing AI spend, usage, and value together in one place across Microsoft and non-Microsoft AI.
Understand AI spend: See what’s driving costs in dollars, where spend is concentrated, find barriers to adoption, and survey employees directly from the app to understand how employees use AI alongside usage data.
Agent 365 Business Analytics cost and forecast page, showing what's driving AI spend in dollars and where spend is concentrated.
Identify the value of AI: Breakdown AI usage into specific use cases to understand what work AI is completing and assess the value of AI through signals such as assisted hours, assisted value, success rate, and satisfaction.
Agent 365 Business Analytics use cases pages, breaking AI usage into specific use cases with signals like assisted hours, assisted value, success rate, and satisfaction.
Turn insights into investment decisions: Once costs and value are connected, make informed decisions to optimize investments with spend forecasts and scenario planning, explore insights in plain language, and share views through Outlook or Teams for deeper collaboration.
Agent 365 Business Analytics customize forecast page, with spend forecasts and scenario planning to help make informed investment decisions.Agent 365 Business Analytics begins rolling out starting in Frontier today. Analytics for Microsoft services require at least one Microsoft 365 Copilot license in the tenant, while analytics for connected non-Microsoft AI services require Agent 365 and/or ME7 licensing.
For more details, please read our deeper blog on Agent 365 Business Analytics and visit our Learn page
Agent 365 Cost Management
In addition to new tools for business leaders, we continue to enhance our core IT tools to control AI spend in Cost Management.
Agent 365 Cost Management, showing what's driving AI spend across users, services, agents, and models, with spending limits and alerts to keep usage within budget.
New Cost Management capabilities help organizations understand what is driving AI spend and put the right guardrails in place as adoption scales:
See what's driving spend. Track consumption across users, services, agents, and models to understand where AI investment is going.
Set the right guardrails. Apply spending limits and alerts at the user, group, organization, or resource level to keep usage aligned with budget.
Match intelligence to the work. Control which models teams can use so organizations can reserve premium intelligence for the work that requires it.
Share controls with delegated access. Let business leaders and their delegates manage their own AI budgets and policies within Cost Management.
Connect AI economics to your existing processes. Use Microsoft Graph APIs, data exports, and Fabric connectors to integrate AI spend into custom reporting, budgeting, and chargeback workflows.
Get started
- Open Agent 365 in the Microsoft 365 admin center.
- Follow the Get started with Agent 365 guide and the training module.
- Bring in Microsoft FastTrack or a Microsoft partner for deployment help.
See Agent 365 at Microsoft Ignite
Join us in San Francisco or online, November 17 to 20.
*Lightning talks and theater sessions are only available to in person attendees. Breakout sessions will be streamed online during the event and available on demand after the event at https://ignite.microsoft.com/.
Learn more
- Agent 365 documentation
- Agent 365 overview
- Microsoft 365 roadmap
- Agents are your newest insiders. Security must move at their speed.
Tell us what you think in the comments, and share what you'd like to see next!