Defender XDR incidents vs alerts: how correlation works and how to triage
Reviewed by ContraForce
Microsoft Security · Tech article · Intermediate · 5 min read
How Microsoft Defender XDR groups alerts into incidents, what the incident queue shows, where attack disruption, AIR and ZAP fit, and why MSSPs should triage incidents, not alerts.
A phishing email lands in a customer tenant. Over the next hour, Defender raises an alert on the message, another on a risky sign-in by the user who clicked, and a third on a new inbox rule in that mailbox. Your analysts can work three alerts, or they can work the one attack those alerts describe.
Microsoft Defender XDR is built for the second option.
What is the difference between an alert and an incident in Defender XDR?
An alert is a single detection from one source. An incident is the container Defender builds around related alerts so you can see the whole attack in one place. Every alert belongs to an incident, and the incident is where your team sets the owner, status and classification.
Alerts can come from the Defender products, Microsoft Sentinel analytics rules and other connected sources. The incident gathers them with the affected users, devices and evidence, so you investigate one story.
How does Defender decide which alerts belong in the same incident?
When an alert arrives, Defender checks whether it is related to other alerts within a time frame. If it is, the alert joins the existing incident. If it is sufficiently unique, Defender opens a new incident. Microsoft does not publish the exact criteria; they are part of its internal correlation logic.
Correlation keeps running after an incident exists. When Defender finds shared entities (users, devices, mailboxes), shared artifacts (files, processes, email senders), overlapping time frames or a sequence that points to a multistage attack, it merges incidents. The contents of one incident move into the other, and the source incident is closed automatically.
When Defender will not merge
Defender skips an automatic merge when one incident is closed, when the two are assigned to different people, when they have different classifications or determinations, when the merge would exceed the entity limit, or, if you enable the condition, when the devices sit in different device groups.
You can merge manually once you fix the blocker, up to five incidents at a time. Learn notes that merging is preferable to moving alerts between incidents, because the activity log is kept.
For customers with Microsoft Sentinel in the Defender portal, each Sentinel workspace is its own data source. Only alerts from the primary workspace are correlated with Defender alerts. Alerts from secondary workspaces are not correlated with Defender or with other workspaces.
What does the incident list show you?
Microsoft now recommends incident cases, found under Cases, then Incident, in the Defender portal. The legacy incident experience is still available. Each incident case gets a priority score: top priority above 85, medium between 15 and 85, and low below 15.
Select a row to read the priority assessment, which explains the score using factors such as severity, threat signals, affected assets, alert types and MITRE techniques. You can filter by severity, status, assignee, classification, detection source, tags, workspaces and more, then save those filters as a filter set.
Where do attack disruption, AIR and ZAP fit in?
All three can act before your analyst opens the incident, so check what they did first.
- Automatic attack disruption: when Defender identifies an active attack such as a ransomware campaign with high confidence, it contains the assets the attacker is using. Actions include contain device, isolate device, disable user, revoke user session and suspend user in Entra. Affected incidents carry an Attack Disruption tag, the incident page shows a yellow banner, and the action is added to the Action center.
- Automated investigation and response (AIR): an alert creates an incident, which can start an automated investigation. Each piece of evidence gets a verdict of malicious, suspicious or no threats found. Depending on your configuration, remediation runs automatically or waits for approval in the Action center. Not every alert starts an investigation.
- Zero-hour auto purge (ZAP): in cloud mailboxes, ZAP acts on phishing, spam and malware messages that were already delivered, searching the last 48 hours of delivered email. Depending on the verdict and policy, it quarantines the message or moves it to Junk Email. Users are not notified.
You stay in control
Your security team can undo any automatic attack disruption action. You can also exclude critical users, devices and IP addresses from automated containment.
Why should you triage at the incident level?
Because the incident is the unit of attack. Attack disruption evaluates the whole attack and acts at the incident level, AIR adds related alerts to a running investigation, and the incident holds the owner and classification. Triage alert by alert and you split one attack across several people.
Splitting also blocks correlation. Defender will not merge related incidents assigned to different people, so two analysts end up investigating one attack in two places.
Back to the phishing example. Suppose Defender places all three alerts in one incident. ZAP may already have pulled the message from other mailboxes, and if Defender judged it an active compromise, attack disruption may have disabled the user. Your analyst checks the Action center and the Investigations tab, confirms what is contained, then handles what remains: credential reset, the inbox rule and anything the user sent while compromised.
What changes when you triage across many customer tenants?
Multitenant management in the Defender portal gives you one view of incidents, alerts and cases across the tenants you manage, plus advanced hunting across tenants. You can organize tenants into named tenant groups and switch the view between them. Each customer Sentinel workspace still has to be onboarded to the Defender portal in its own tenant.
Agree one convention for ownership and classification and apply it in every customer tenant, so an assigned, merged or closed incident means the same thing everywhere.
How should your analysts work the queue?
- Sort by priority score and read the priority assessment, rather than relying on severity alone.
- Open incidents tagged Attack Disruption first. Containment is already in place, so review the Action center before you contain the same user or device again, and agree with the customer before you undo anything.
- Check the Action center for AIR remediation waiting for approval before you start manual work.
- Before you search and purge email by hand, filter Threat Explorer on the Additional action column for ZAP. ZAP is not logged in Exchange mailbox audit logs.
- Give each incident one owner. Assigning related incidents to different analysts, or classifying them differently, stops Defender from merging them.
- Save a filter set per tenant group or workspace, so each shift starts from the same view.
If you deliver this service through ContraForce, the Security Delivery Agent re-checks the live Defender incident before it starts and skips incidents that AIR already resolved or that were merged into another incident. Attack disruption incidents are never delayed. ContraForce does not yet check ZAP or containment state when a Gamebook action runs, so the Action center stays your source of truth. Details are in Defender AIR, ZAP, and Attack Disruption.
Keep learning
- Configuring Security Delivery Agents: On Queue, policies and shifts. Set which severities a Security Delivery Agent picks up On Queue, gate its Gamebooks with a confidence threshold, add investigation policies and shifts, link SOPs and check Run History.
- Writing incident response SOPs that analysts and AI agents can both follow. A structure for incident response SOPs that analysts and AI agents can follow the same way: decisions as explicit conditions, per-customer exceptions, versioning and a suspicious sign-in example.
- Controlling Microsoft Sentinel costs across customer workspaces. How to keep Microsoft Sentinel spend under control in each customer workspace: table tiers, commitment tiers, retention, ingestion-time filtering, free data sources and usage monitoring.
Sources
- Microsoft Learn, Alert correlation and incident merging in the Microsoft Defender portal, verified 2026-11-20
- Microsoft Learn, Prioritize incident cases in the Microsoft Defender portal, verified 2026-11-20
- Microsoft Learn, Automatic attack disruption in Microsoft Defender, verified 2026-11-20
- Microsoft Learn, Automated investigation and response in Microsoft Defender XDR, verified 2026-11-20
- Microsoft Learn, Zero-hour auto purge (ZAP) in Microsoft Defender for Office 365, verified 2026-11-20
- Microsoft Learn, Microsoft Defender multitenant management, verified 2026-11-20