AI decision models are the new kids on the block, with TypeSafe AI’s Jev,1 Cloudflare’s Clef,2 and Microsoft-Decision-1.3 Frontier models are powerful for complex analysis and generating text, but automation often needs smaller answers: which category, which queue, or whether to escalate. Decision models focus on these bounded tasks, returning structured choices and probabilities without generating free-form text. This can reduce latency and cost while giving AI-based workflows outputs they can act on directly. Probability thresholds can then determine whether to proceed, request another evaluation, or hand the decision to a human in the loop.
In short:
An AI decision model reads an input and returns an answer from a set you define (an option, a score, or the probability that something is true), with a calibrated probability for every allowed answer, in one pass and without generating text. - System One Models4
So I was curious to test the Microsoft-Decision-1 model and better understand how it works in an experiment related to SecOps and incident automation.
My Experiment # When alert volumes are high, SOC teams tend to prioritize medium- and high-severity alerts and use automation to triage or close lower-severity alerts. I tested whether Microsoft-Decision-1 could identify alerts worth escalating or recommend a lower severity based on the alert and included alert evidence. These outcomes could support incident reviews by highlighting alerts that are worth another look or support detection engineering to adjust a detection’s severity level.
Gartner describes ISOC (Integrated Security Operations Center) as a converged technology approach that combines threat detection, investigation, and response (TDIR) in a single-vendor suite or cloud service.1 For Microsoft, that means Defender XDR and Microsoft Sentinel. This shift started a while ago with the move towards the unified security operations platform (USOP).
With ISOC, Microsoft 365 E5/E7 customers without Sentinel can now use features previously available only to customers using both Defender XDR and Microsoft Sentinel.
This post covers the ISOC in Microsoft Defender preview. Capabilities and availability may change during the preview. See the official ISOC in Microsoft Defender documentation and Microsoft’s announcement. Why Microsoft is doing this # You could already be a Sentinel customer without paying for the SIEM features: Sentinel lets you ingest some first-party Microsoft data, mainly alerts, as part of free data sources.
In my view, Microsoft is following the industry’s direction while also helping shape it. ISOC is now its own category in Gartner’s 2026 Hype Cycle for Security Operations, alongside security data lakes.2 Both are newer alternatives to a classic SIEM.
When speaking with customers and colleagues, I often encounter questions and misconceptions about Defender for Servers in Microsoft Defender for Cloud. In this post, I explore these questions:
Why should I care about Defender for Servers? How should I onboard my servers to Microsoft Defender for Servers? What happens under the hood when I onboard servers? How can I selectively deploy Defender for Servers (e.g., for a pilot)? How can I enforce and maintain full coverage of Defender for Servers? How can I monitor and query Defender for Servers coverage? Typical confusion of Defender for Servers Defender for Servers Plans and Licensing # While client devices in Microsoft Defender for Endpoint (MDE) are usually licensed through Microsoft 365 subscriptions such as Microsoft 365 E5, servers do not benefit from this licensing model. Instead, server licensing follows the Azure consumption model.
If you work with the Microsoft account team, you might also be able to obtain Microsoft Defender for Endpoint Server licenses. These are not consumption-based, so they do not require an Azure subscription and are usually part of enterprise agreements.
Uuuh, a phishing email, but actually a good one! That’s what caught my attention and led me to conduct a mini investigation into a device code phishing campaign in the wild, which turned out to be a recent version of Kali365.
Device Code Phishing Recap The attacker first starts a legitimate device code authentication request for a legitimate Microsoft first-party application. The victim is then tricked into entering the attacker’s device code on Microsoft’s real sign-in page and completing authentication, including MFA. Once approved, Entra ID issues the tokens to the attacker’s waiting session.
During a recent engagement I stumbled across an interesting observation: the Defender for Cloud Apps advanced hunting table CloudAppEvents was empty, even though the tenant had adopted Microsoft 365 collaboration and security tooling.
No events in the CloudAppEvents table The Microsoft 365 app connector was reported as healthy, with the initial connection dating back to 2022:
Healthy App Connector However, when opening the connector, none of the Microsoft 365 components were actually selected:
No components selected So either someone forgot to tick the boxes back in 2022, or Microsoft used different auto-provisioning defaults at the time.
This is also called out in the Microsoft docs 1:
In January 2026, the default values were added to support complete security coverage. If you configured your application before January 2026, make sure you select all of the default options and select Connect again to update your configuration.1
For maximum protection, we recommend selecting all Microsoft 365 components. Some threat detection and response functionalities don’t work unless all required components are properly selected.1
Interestingly, on another, more recently onboarded tenant where the connector had never been modified, all boxes were ticked except for file monitoring.
