Microsoft Ignite 2026: An Identity and Firewall for Every AI Agent

· Cortex Solutions

An enterprise control room with a large display showing a network of glowing AI agent nodes, each protected by a shield, linked to cloud servers through a filtering firewall layer

At Ignite 2026 Microsoft expanded Entra Agent ID to discover 60 local agent types, auto-disable idle agents and add an MCP firewall, plus new Azure ops agents.

Microsoft Ignite 2026 opened on November 17 in San Francisco and online, running until November 20 according to Microsoft's official Book of News page. Many of this year's announcements centre on AI agents in the enterprise, and the clearest message among them is this: how to know which agents you have, give them the right permissions, and switch off the unused ones before they become a security gap.

This article relies on two sources: the Ignite 2026 Book of News page on news.microsoft.com, and a Virtualization Review article by David Ramel published on November 17, 2026, titled "Microsoft Ignite 2026: Top Azure Announcements for Enterprise IT".

An identity for every agent: Microsoft Entra Agent ID

The most striking theme this year is treating AI agents like digital employees, each with an identity and permissions. According to Virtualization Review:

  • Microsoft Entra Agent ID can now, in preview, discover local agents running on employees' machines, supporting discovery of 60 local agent types, including OpenAI Codex, Claude Code and GitHub Copilot CLI.
  • Automated lifecycle management for agent identities is now generally available (GA), with a default policy that disables identities that have been inactive for 60 days.
  • Microsoft introduced, in preview, dynamic agent risk assessment and Conditional Access policies that distinguish agents from users.
  • It also announced an "MCP firewall" in preview; MCP is the protocol through which agents connect to tools and data.

The practical meaning is simple: many employees today run coding tools and AI assistants on their machines without the IT department knowing, and the new direction is for these tools to be visible and managed just as human accounts are.

Agents that run the cloud itself

The second group of announcements concerns agents working inside Azure to help operations teams. According to the two sources:

  • The Azure Resource Manager MCP Server, which previously allowed reads across more than 3,500 Azure resource types, now supports creating, managing and updating resources, subject to Azure Policy.
  • The Azure Copilot deployment agent, which turns natural-language requirements into infrastructure and deployment code, is generally available according to Virtualization Review.
  • Azure SRE Agent received preview enhancements, including customer-managed encryption keys, Entra Agent ID integration and cross-agent communication.
  • Azure Monitor health models are now generally available.

Modernisation and cloud migration: Azure Catalyst

Microsoft presented Azure Catalyst as a unified modernisation offering that combines Azure Migrate and GitHub Copilot modernisation capabilities with AI agents. It covers VMware and Hyper-V discovery, replication and migration tracking, and assessment extends to containers, SQL, MongoDB, Oracle and file servers. According to the Book of News, Azure Catalyst is generally available, with new capabilities in preview.

Other notable announcements

  • Microsoft announced that Azure Cobalt 200 virtual machines are generally available, saying they deliver up to 50% better CPU performance than Cobalt 100, with plans to make them available in 24 Azure regions by year-end.
  • Windows Server 2028 is now in preview.
  • Microsoft Perception, previously known as Project Perception, is now generally available, and Microsoft said its expanded agents cover about 80% of typical Defender alerts.

It is worth noting that many of these capabilities are still in preview and not yet ready for full production, and that availability details may vary by region.

What does this mean for Saudi businesses?

This section is Cortex Solutions' own reading of the announcements, not a statement attributed to Microsoft:

  • Inventory your agents before you scale: know which AI tools your employees actually use, on which devices and with which permissions. You cannot protect what you don't know about.
  • Treat an agent like an employee: every agent should have an identity, an owner, least-privilege permissions and a review date. Disabling inactive identities is an idea you can apply on any platform you use.
  • Set limits on what an agent can write: once an agent can create and modify resources rather than only read them, its actions should pass through clear policies, with human approval for sensitive steps.
  • Don't build on previews alone: test new capabilities in a trial environment and wait for general availability before relying on them in critical systems.
  • Tie it to your local obligations: any agent that reaches customer or employee data should be reviewed against the data-protection requirements in force in the Kingdom and your company's own policies.

At Cortex Solutions we help companies build AI agents and automation for their business with clear permission controls and human review. Learn about the service:

https://cortex.sa/services/ai-automation

Read next