Like an Archaeologist digging through the book of news, attending hidden sessions, checking out the labs, and navigating around the sudden waterfalls that appeared at The Hub in Moscone South, my aim is simple:
“Find out what new Security news was released this week at Ignite so far.”
I, among many others, descended upon the rolling hills of San Francisco this week (or tuned in for the livestreams) to get a first look at all the new releases and updates Microsoft is rolling out across its ecosystems. If you attended or watched the Keynote, one thing became abundantly clear:
AI is the center of the Universe at Microsoft, and everything is being built to either leverage or support it.
If you are an AI engineer, Modern Work advocate, productivity nerd, or even a hardware junkie, the conversations at the Keynote and beyond have been incredibly rewarding. Copilot this, AI model that, Foundry gets more stuff, new Silicon for AI, “Copilot is the new Work OS”, and so on and so forth. If you’re a security practitioner like myself, however, you may have felt a little left out and asking questions like:
Where is all the major Agent 365 news?
What’s new in Purview, Entra, and Defender?
How are we securing these new AI-driven ecosystems?
CAN WE PLEASE STOP SAYING COPILOT?
Have no fear, my friend, as there was still some content meant for this of us who man the walls guarding our organizations from risk outside and in. It just wasn’t in your face. If you have read the Book of News already, you will be familiar with much of this. If you haven’t, I recommend you check it out for more in-depth summaries and links to official sources. As for this article, I want to highlight a few items that I found particularly interesting or important and see if I can bring some additional awareness to them and hopefully provide an extra resource or two!
Gabe’s Favorite Security News from Ignite 2026

Agent 365
A few things stood out to me here that I want to touch on.
Dynamic Agent Risk (Preview) lands in Agent 365, allowing defenders to pull in signals from across the security estate to establish a risk profile for an agent without having to manually correlate these signals together. This will change (hence dynamic) throughout the agent’s lifecycle. I think this is incredibly valuable as we start to expand our definition of risk outside of the User sphere and into Agents. Agents aren’t themselves trying to introduce risk to your organization. They are trying to solve problems provide by the user. It’s the tools they call, actions they attempt, and data they consume that introduces risk into an Agentic system. Having a dynamic risk register that follows the lifecycle of the Agent will give us stronger indicators in where we need to spend time implementing guardrails or reducing unnecessary risk so our Agents behave within the bounds we set for them.
Expanded Governance for MCP servers as well as expanded Local and Shadow AI Governance capabilities. You can now (in preview) integrate MCP servers from Azure AI Gateway and LiteLLM directly into Agent 365 for better observability. What agents are using them, how, for what purposes and control their availability. For our Local and Shadow AI, we are getting more granular controls than just blocking as well as getting visibility into “Shadow MCP” servers. The shadows, they haunt us.

Defender
While the new ISOC core capabilities that just went generally available are great, one particular item caught my eye.
Extending runtime protection for local agents is a huge win. Now we can ensure our use of local AI agent tools like Claude Code, Codex, Copilot, and more can be secured by monitoring the agent loop and blocking malicious instructions before these agents can act. Couple this with Network DLP via Entra and Purview and you can extend this protection even further by ensuring our sensitive information doesn’t leave over the network. I got to see a little of this in action at a demo and a lab, and this to me is the crown jewel. Defender is a product so many of us have already deployed and use, now giving us even more value via runtime protection for local agents!

Entra
Entra some some nice new expansions get introduced this week that are worth highlighting.
Entra Agent IDs for Local Agents and Autopilots. Now as we discover and use local agents, we can assign Agent IDs to them so we can differentiate between user and agent activity, and also give our Autopilots specific Agent IDs so we can secure and monitor their access to protected resources. The Autopilot announcement caught me a little off guard as I would have assumed this was happening already, but perhaps not? I must admit my knowledge of Autopilots is slim so perhaps this is due to the new Copilot Workspaces feature. The local agents one makes total sense and I’m happy to see local agents get Agent IDs now.
Expanded PIM capabilities came to Entra this week and a few things really stood out. For starters, users can now activate roles and access Azure VMs and Azure ARC VMs from the My Access portal. Users now have a centralized location to get their roles, VMs, access packages, manage the agents they own/manage/sponsor, as well as handle their own security info. Additionally, Agents now have the ability to activate and deactivate assigned roles, preventing the need for them to have long-standing permanent credentials. This does (from what I can see) require the agent to activate their role via the Azure Resource Manager API (little more of a pro-code approach) but it’s a step in a better direction. Last, but definitely not least, is the ability to active App roles in PIM. To keep it succinct, app owners can now define the roles that an application uses to make authorization decisions and bring them under management with PIM. Users can then activate an App Role to get the desire authorization into an application, and simply deactivate when they are done, so app owners don’t need to hardcode authorization values or use other values or transforms to determine what role needs to be present in token.
Entra Global Secure Access saw some nice improvements and got some good demo attention this week in the Agentic Security conversation. Agent Traffic Acquisition, MCP Firewall, and network visibility into Local Agents are all nice to see. Couple GSA with Defender, Entra, Agent 365 and Purview, and we now have a full-stack security suite for the Agentic Enterprise! Granted, you will want the E7 to get your hands on all these features together, but for a Microsoft-native organization the benefits are starting to really add up against the $99/month/user E7 bill.

Purview
Purview’s announcements were a little light, but saw some major improvements rather than new features.
Faster time to protection by increasing the number of files daily that can be autolabeled from 100k to 500k. This is a huge jump in capability that will help organization close the gap on labeled data faster, but it’s only one part of the equation.
Improving Copilot adoption by grounding Copilot protection in Sensitive Information Types present in a file even if the file itself doesn’t carry a sensitivity label. This removes one of the main Copilot adoption hurdles by giving organizations reduced risk of sensitive data exposure while they wait for Purview to continue autolabeling their data.
We also heard from Microsoft this week that they are taking a big initiative in improving their out-of-the-box Sensitive Information Types to greatly improve their accuracy. Their stance was clear: “If we provide you this SIT, you should be confident that it works, works well, and in the right context in which the data was present.” (I’m paraphrasing).
This was just a taste
The messaging from Microsoft this week was clear. These AI capabilities are here to fundamentally change the way we work, and the Security needed to protect our organizations is coming with it. My post didn’t include some of the major updates for Intune (Server management, woo!) or Project Perception, Windows, or any of the Azure services, as there was too much to include here. I highly recommend again taking a look at the book of news and reviewing the Breakout session recordings for anything you might have missed. It’s going to be a big year for Microsoft Security projects and initiatives. Now go out there and Secure those Agents!



