A security analyst monitors AI robots managing cloud servers, databases, and cybersecurity checks in a neon-lit control room.
Microsoft opened Ignite 2026 on November 17 with a pile of Azure announcements. Taken together, they point one way: AI agents are moving from reading your cloud estate to changing it. Azure's identity, policy and governance layers are meant to keep those agents on a leash.

Five announcement areas matter most for enterprise IT. Some are generally available and some are previews, so that label matters a lot for anyone planning a pilot.

A security analyst monitors AI robots managing cloud servers, databases, and cybersecurity checks in a neon-lit control room. Agentic operations: ARM MCP Server, Copilot, SRE Agent and Monitor​

Microsoft's Brendan Burns framed the push as two jobs: operating agents, and having agents help operate. Teams can connect their own agents through Model Context Protocol (MCP), use Azure's built-in agents, or extend Azure SRE Agent to external systems. Microsoft says all three paths build on Azure's existing identity, permissions and policy controls.

What changed in the ARM MCP Server

  • The server previously gave MCP-compatible agents read access to Azure resources through the control plane, across more than 3,500 resource types. Agents can now also create, manage and update resources.
  • Microsoft says requests are evaluated by Azure Policy. Policies that block noncompliant deployments, require tags, or restrict regions and SKUs apply whether a change comes from the portal or from an agent.
  • Microsoft's product team describes the pattern as discover, plan, act and verify. Tools cover resource and resource-group create/update, resource actions such as restart or key retrieval, and deployments with a What-If preview. They also cover deployment status and cancellation, single-resource reads, and cost queries.
  • Pairing the ARM MCP server with the Bicep MCP server lets you author Bicep, build it to ARM, then deploy and manage through ARM MCP.
  • A Skills feature is experimental and off by default. Clients must opt in by sending the request header x-mcp-toolset: Skills. Microsoft warns that tool names and behavior may change.

Copilot, SRE Agent and Monitor

  • The Azure Copilot deployment agent is generally available. It turns natural-language requirements into infrastructure and deployment code.
  • Copilot spaces and custom skills are in public preview. Azure Workload Center, which extends agentic operations to Oracle and SAP, is in limited preview.
  • New SRE Agent preview features are customer-managed keys, Entra Agent ID, Agent 365 integration and cross-agent communication. Teams integration covers investigations across multiple clouds.
  • Azure Monitor health models are generally available. They connect telemetry to architecture, dependencies and business impact.

Reality check. Policy enforcement is a real guardrail. Azure Policy still only blocks what your policies describe, though, and a compliant but wrong change will sail through. Test with What-If, scope permissions tightly, and don't assume policy replaces human review. The customer figures Microsoft cites, such as Ecolab's reported 60% faster root-cause identification, come from Microsoft's own blog. They are not independent benchmarks.

Azure Catalyst: migration and modernization under one roof​

Azure Catalyst is Microsoft's new unified, agent-led modernization offering. It brings Azure Migrate and GitHub Copilot modernization together and extends both. Microsoft's Azure documentation is already being rebranded to match. An Azure Local article now describes Azure Catalyst as "formerly Azure Migrate." Third-party coverage also says existing Azure Migrate projects continue to work. That claim comes from a secondary source, so confirm it in Microsoft's documentation before relying on it.

What's in it

  • Discovery that links infrastructure with application code, so teams can trace dependencies between servers, databases and applications.
  • Modernization strategy, cost comparisons and AI-generated migration waves.
  • For VMware and Hyper-V, discovery, replication and tracking with Azure Local as a destination, while preserving supported Azure Arc settings.
  • Direct migration to Trusted Launch is generally available for eligible VMs.
  • SQL migrations organized into waves, with execution runbooks integrated with Azure Database Migration Service.
  • Wider assessment for containers (AKS or Azure Red Hat OpenShift), MongoDB, Oracle, file servers and NAS, per Redmondmag.
  • GitHub Copilot assessment and transformation for .NET and Java apps alongside their databases. Governance rulebooks and bring-your-own-model support are generally available.
  • Security analysis that flags vulnerable code and databases. Microsoft says GitHub Copilot can then help remediate them as apps are modernized.

Caveats. Microsoft's Catalyst post says the new capabilities are public preview unless noted as generally available. Many expanded workload features are therefore not production-ready. The "1.4 million applications modernized" figure and the speed claims are Microsoft's own, with no independent verification. Treat them as marketing until you've piloted the tool on your own estate. One preview detail will matter to some readers. Migrating VMware and Hyper-V VMs to Azure Local in Azure Government through Catalyst is a preview.

Azure Arc and Machine Configuration: baselines that fix themselves​

For sysadmins, this is the most practical item. Azure Machine Configuration adds a portal hub, centralized inventory and, in public preview, security baseline remediation.

  • Hub. A new Machine Configuration hub in the Azure portal covers enrollment, inventory, compliance and governance. Its "Get Started" tab walks through enrolling a first machine, and Enablement can deploy prerequisites across an entire scope.
  • Inventory. The Windows-focused experience covers applications, certificates, services, users, groups and Windows roles. Linux covers daemons and software. You can inspect one machine or query the fleet with Azure Resource Graph.
  • Remediation. Previously you could assess machines against customized baselines. Now you can pick supported settings for Machine Configuration to actively maintain, and it corrects drift. Settings left in audit mode only report.

Microsoft's availability table:

BaselineEffects
Azure Security Baseline, LinuxAudit, Remediate
Azure Security Baseline, WindowsAudit, Remediate
CIS LinuxAudit, Remediate
CIS Windows ServerAudit only
STIG LinuxAudit, Remediate

Multicloud. Servers discovered and onboarded from AWS and Google Cloud through Azure Arc Multicloud Connector can be enrolled in Machine Configuration and assigned baselines from the start.

Windows recovery. Arc-enabled servers can now have Windows Recovery Environment audited and enabled centrally. Arc can also configure Quick Machine Recovery, which Microsoft says brings automatic remediation for eligible startup failures on Windows Server 2025. That is limited to eligible failures on that version, not a blanket recovery promise.

Practical advice. Auto-remediation is powerful and can also break things. Start with audit mode, review what would change, and enable remediation on a small pilot group before widening it. Remember that remediation is still preview, so check your support expectations.

Entra: giving AI agents their own identities​

Microsoft's central argument is that agents should be governed as distinct identities, not as extensions of human permissions. Microsoft cites its own telemetry that tool and agent abuse accounted for 12% of observed attack activity on Azure AI workloads.

  • Discovery (preview). With Defender and Entra Global Secure Access, Entra can discover 60 local agent types, including Claude Code, OpenAI Codex and GitHub Copilot CLI. Each discovered agent receives an Agent ID.
  • Lifecycle (generally available). Policies check for a human sponsor, whether the agent is still needed, and recent activity. Entra can notify, then disable or delete according to your settings. The out-of-the-box policy disables agent identities inactive for 60 days.
  • Conditional Access (preview). Policies can tell people apart from local agents acting for them, and can restrict sensitive actions during risky sessions. Conditional Access for Windows 365 for Agents is generally available.
  • Dynamic Agent Risk (Agent 365 preview). It combines signals from Entra, Defender and Purview into a continuously updated risk score.
  • MCP firewall (preview). It enforces policy across MCP servers, tools, resources, methods and protocol versions.
  • Privileged Identity Management (preview). Just-in-time access to Azure and Azure Arc VMs, with privileged session monitoring and recordings.

The pattern is familiar to anyone who has done identity governance. You discover, assign an owner, apply least privilege and retire what's idle. Most of the headline agent-risk controls are previews, though. Don't expect to roll them out tenant-wide tomorrow.

Foundry IQ: knowledge graphs and a retrieval model​

Microsoft is grouping its context features under a new "Microsoft IQ" umbrella. Foundry IQ's pieces are:

  • Knowledge graphs (preview). They let agents reason across relationships in organizational knowledge, not just keyword results.
  • Microsoft Retriever 1 (preview). A retrieval model for agentic grounding. Redmondmag reports it plans retrieval against existing indexes to cut searches and cost. That is the trade press's reading, so test the savings yourself.
  • SharePoint indexed knowledge source (generally available). It supports permissions, so agents respect users' Entra group and SharePoint site access.
  • Foundry IQ Serverless (December 2026). Microsoft says it will be generally available with usage-based pricing and no capacity to provision. Free monthly allowances are planned. No exact date or price is given.

Permission-aware retrieval is the item to scrutinize first. An agent that surfaces a document the user shouldn't see is a data leak, and clean SharePoint permissions are a prerequisite.

What to do now​

  1. Sort by availability. Generally available items include the Copilot deployment agent, Monitor health models, Entra lifecycle policies, the SharePoint knowledge source and Trusted Launch direct migration. Most of the rest is preview or experimental.
  2. Pilot read and What-If first with the ARM MCP server, and confirm which Azure Policy assignments actually cover agent-driven changes.
  3. Run Machine Configuration baselines in audit mode before turning on remediation.
  4. Inventory your local AI agents. Developers' CLI agents are likely already on managed devices.
  5. Check licensing and tenant availability. Several features depend on Agent 365, Defender or Global Secure Access.

The bottom line​

None of this is a single blockbuster. It is a steady build-out of a thesis: agents will manage infrastructure, so give them identities, policy limits and audit trails. The governance story is credible on paper. The caution is that so much of it is preview. Microsoft's own post says the real test is how it performs in your environment, and that is fair advice.

 

References

  1. Major Azure Updates for Enterprise IT Land at Microsoft Ignite - Redmondmag.com Redmondmag.com Tue, 17 Nov 2026 22:01:58 GMT
  2. Your Privacy Choices Opt-Out Icon azure.microsoft.com
  3. Your Privacy Choices Opt-Out Icon azure.microsoft.com