Blog Post

Microsoft Foundry Blog
4 MIN READ

Govern your AI fleet with confidence on Microsoft Foundry

ronglums's avatar
ronglums
Icon for Microsoft rankMicrosoft
Nov 18, 2026

The AI era is lowering the barrier to building agents—teams can go from idea to working prototype faster than ever. But taking agents into production across an enterprise is where the real challenge begins. As agents spread across teams, projects, and subscriptions, organizations need to answer harder questions:

  • Which agents, models, and tools are running across our organization?
  • Is this agent allowed to call that tool?
  • Why did spending jump last week, and which project caused it?
  • Are we compliant with the policies we've set, and where are the gaps?

That's where Microsoft Foundry stands apart: governance is built into the platform, not bolted on after agents reach production. Platform teams set clear organizational boundaries, and developers build and deploy with confidence within them. In this post, we'll explore new Foundry governance capabilities that help organizations build, operate, and govern AI agents at enterprise scale.

Microsoft Foundry: A strong foundation for governed AI

Foundry governance connects the controls platform teams need with the flexibility developers need: set the boundaries centrally, then let teams build within them. Instead of stitching together separate tools after agents ship, platform teams define requirements once—through Azure Policy and Foundry's own controls—and apply them across the subscriptions and projects in scope. Developers keep moving fast, and every agent they build starts inside an approved, observable, and accountable boundary.

Foundry provides built-in governance for agents, models, and tools across four areas:

  • Agent identity: Foundry gives every agent an identity of its own. Represent agents with Microsoft Entra agent identities and blueprints, so they can be discovered, managed, and audited like other identities in your organization. Use supported authentication and on-behalf-of-user (OBO) delegation so agents reach downstream resources with the right permissions—their own, or the signed-in user's.

  • Agent access: Decide which models, agents, tools, and network destinations are allowed, using Azure Policy and Foundry controls at the scope you choose:

    • Models: Model deployment policies govern which models are approved and eligible for deployment, so teams build on models your organization has vetted.
    • Agents: Agent deployment policies restrict which agent kinds can be created and which container images hosted agents can run.
    • Tools: Native tool governance uses Azure Policy to check tool configurations, such as approved tool types and MCP endpoints, while project-level controls govern runtime tool access.
    • Network egress: Hosted-agent egress policies check supported outbound network configurations, with runtime enforcement through egress guardrails.
  • Cost management and FinOps: See where spend is going and stay ahead of it. Monitor costs by agent and model, manage model quotas, and configure cost controls (preview) with spending or token budgets that alert—or block usage—when thresholds are reached.

  • Safety and responsible AI: Configure guardrails and runtime controls that detect and mitigate risks in user inputs, model outputs, and agent interactions, so safety travels with every agent.

Together, these capabilities help teams answer four essential questions: who is this agent, what can it access, how is its usage managed, and which safety controls apply?

See your AI fleet at a glance

Open Manage in the Foundry portal to get a cross-project view of your AI estate and the signals that matter:

 

  • Fleet: Explore inventory, activity, token consumption, cost, and policy-violation highlights.
  • Cost: Analyze spending and forecasts, and manage cost controls.
  • Operations: Review invocations, errors, latency, token usage, and agent quality.
  • Compliance: Review policies, asset compliance, Defender recommendations, and Purview insights.
  • Quota: Review model quota and request increases.
  • Gateways: View and connect gateways with Foundry Project to centrally manage traffic to models and tools.

Fleet, Cost, and Operations let you start at subscription scope, then narrow by resource group or project. That means you can spot patterns across your estate and follow them into the project that needs attention.

Extend governance with the services you need

Foundry's built-in controls provide a strong starting point. For requirements that call for additional capabilities, connect services you may already use:

  • AI Gateway: Centralized traffic controls for model and tool endpoints, including routing, rate or token limits, and monitoring. These apply to traffic routed through the gateway.
  • Microsoft Agent 365: Enterprise agent inventory, identity-based governance, observability, and lifecycle management across supported platforms.
  • Microsoft Purview: Data security and governance, including audit, data protection insights, and supported data loss prevention scenarios.
  • Microsoft Defender: Security posture, threat protection, and investigation and response.

These integrations extend the foundation; they're not prerequisites for Foundry's native identity, access, cost, or safety controls, and they aren't interchangeable control planes. 

Put governance into practice

Start with Foundry's built-in controls, then use fleet-wide signals to guide investigation and action:

  1. Set identity and access boundaries. Review agent identity and the identity and access overview, and grant only the permissions your scenario requires.
  2. Apply policies and safety controls. Configure applicable model and agent deployment policies, tool governance, and guardrails. Verify configuration checks and runtime behavior separately.
  3. Establish cost visibility. Check cost permissions, review subscription costs, and configure threshold actions.
  4. Monitor and investigate. Use Manage to spot signals in Fleet, then follow them into Operations, Cost, or Compliance. For example, investigate an agent's latency in Operations and continue to project-level traces where available.
  5. Extend where needed. Add AI Gateway, Agent 365, Purview, or Defender when your requirements call for their capabilities.

With governance built into Foundry, platform teams can set clear boundaries and gain fleet-wide visibility while developers keep moving from ideas to production. Build boldly—and govern with confidence. Microsoft Foundry brings agent development and enterprise governance together, so teams can move from ideas to production with the boundaries and visibility they need. Install the Foundry Dev Pack to start building and learn more about Foundry governance.

Updated Nov 18, 2026
Version 2.0
No CommentsBe the first to comment