Microsoft SC-500: Incident Response Across Microsoft Security
A unified Microsoft Defender incident view can connect identity, endpoint, cloud and Sentinel signals, but containment succeeds only when investigators establish scope, preserve evidence and assign safe actions across the right owners. The goal is to determine what happened, which identities and assets were affected, and what containment can be performed without destroying useful evidence. Cross-product dashboards assist that work only when correlation, permissions and investigation ownership are understood.
On this page
- Investigate incidents with specific hypotheses
- Use entity relationships to define the investigation boundary
- Combine Defender XDR and Sentinel evidence
- Hunt beyond the alerted slice of activity
- Block the next malicious action
- Automate deterministic enrichment before disruptive response
- Preserve evidence and restore trust after containment
- Document uncertainty and handoff decisions during response
- Turn incident lessons into durable engineering changes
Begin with the incident hypothesis, not the alert count
An incident aggregates related detections so investigators can reason about one possible attack story rather than chase every product alert independently. Correlation is helpful but imperfect: several alerts might be one compromise, or an incident may omit activity from an unconnected cloud or SaaS source. Start by asking what the attacker might be trying to accomplish, which entities are affected and which evidence would confirm or reject the hypothesis. Assign an incident coordinator to preserve a shared timeline while specialized endpoint, identity, email, data and platform teams investigate their own controls.
Consider an unusual sign-in followed by a suspicious device process and access to an Azure resource. List the expected sequence and test it against event timestamps instead of assuming the Defender incident title establishes causation. A benign software deployment could explain the process while the risky sign-in is unrelated. Record confidence and source of each link, then expand only when the evidence supports it. Correctly rejecting a false relationship is as valuable as finding a real lateral movement path.
Use entity relationships to define the investigation boundary
Accounts, devices, applications, IP addresses, files, mailboxes and cloud resources are the pivots that link otherwise separate signals. An IP address may be shared by many users, while an application object ID or device identifier can provide a narrower correlation key. Validate join assumptions before concluding that every alert involving one IP belongs to the same actor. Resource context also matters: one suspicious call to a production Key Vault may deserve higher priority than several failed attempts against a disposable test environment.
Make an entity worksheet for the incident containing first-seen time, suspected role, affected asset, supporting event and next query. If a device is associated with an identity, confirm the sign-in and process timeline rather than treating co-occurrence as proof of compromise. Ask the Azure owner whether the principal could retrieve sensitive data, and ask the application owner whether the API call matches expected work. This produces a bounded response scope and avoids indiscriminate containment of unrelated services.
Combine Defender XDR and Sentinel without assuming identical retention
The Microsoft Defender portal can expose Defender XDR and onboarded Sentinel signals in a common operating surface. Defender’s native hunting data, Sentinel Analytics-tier tables and longer-retention Data Lake data do not necessarily offer the same query performance, features or historical window. The incident team must know which connector provided a record, how fresh it is and whether it remains available when the investigation broadens backward in time. A unified visual portal should not be mistaken for a single universal collection system.
Before a major incident, document which identities, servers, network devices and custom applications contribute events and which Sentinel tables hold them. During a simulated case, find the same identity across sign-in and workload logs and compare timestamps. If one expected source is silent, create a telemetry gap rather than declaring the behavior never occurred. Confirm that investigation roles can access relevant workspaces without unnecessarily granting high-impact containment privileges.
Hunt beyond the alerted slice of activity
Alerts often identify one known malicious event, not the complete campaign. Investigators may need to search for the same command, hash, domain, OAuth permission or behavior across additional devices and users. Begin with a specific extension hypothesis and use KQL to filter by time, entity and signal, then expand deliberately. A query returning thousands of events is not automatically stronger evidence than a carefully interpreted small set. Threat hunting should distinguish related behavior from routine systems producing similar benign events.
For a simulated compromised account, look for additional app-consent grants, uncommon resource access and corresponding endpoint actions within a defined window. Validate the identifiers and sample raw events before concluding that one user accessed all the resources listed in a join result. Preserve exact query text, time filters and workspace context so another responder can reproduce the finding. If the hunt reveals an unprotected log source, capture that as a separate engineering gap for after containment.
Contain the path that enables the next malicious action
Containment may require disabling or restricting an account, revoking sessions, isolating a device, blocking an indicator, changing firewall rules or suspending an application identity. The correct action depends on the attacker foothold and the business consequence of disruption. Isolating one laptop is inadequate if an attacker still controls a service principal with broad cloud rights. Conversely, disabling an entire subscription based on one low-confidence alert may cause avoidable business damage. Human approval and narrower scope are especially important for high-impact actions.
Write a containment matrix mapping suspected foothold, approved responder, action, expected blast radius and rollback conditions. Test one benign simulated compromise in an isolated environment: revoke the test identity’s access and verify a previously permitted request fails, while unaffected users continue operating. Capture the exact event time and approval. A mature playbook makes the emergency choice understandable without requiring analysts to improvise credentials or guess which service will break.
Automate deterministic enrichment before disruptive response
Sentinel automation rules and Logic Apps playbooks can enrich entities, route notifications and create tickets, while Defender automations may perform supported investigative or response tasks. Start with predictable, low-impact operations that have clear success and failure criteria. An enrichment action that retrieves asset owner or business sensitivity usually carries less risk than an automation that isolates servers or disables privileged accounts. Idempotence matters: a duplicate incident update should not create repeated destructive actions or conflicting tickets.
Design a playbook with a test incident, exact trigger, safe entity lookup and one notification destination. Verify duplicate events, timeout behavior and permission denial. Record which responder reviews the result and which actions require explicit approval. If a playbook cannot find the asset owner, it should surface that uncertainty instead of making an irreversible assumption. Automation should shorten the time to an informed human decision rather than silently replacing it with an unreviewed action.
Preserve evidence and restore trust after containment
Recovery is more than closing the visible alert. The team needs a defensible event timeline, probable entry point, affected identities and data, evidence preservation and a plan to remove persistence. Reimaging a device may not revoke stolen tokens. Resetting a password may not remove an unauthorized app consent. Changing a network rule may block one route while an overprivileged managed identity remains active. Restore the actual trust boundary affected and verify that the necessary business service can safely resume.
Keep a restoration checklist listing sessions, credentials, application grants, device state, workload images, network paths and protected backups as appropriate to the incident. For each item, name its owner and direct verification. Document the evidence collected before disruptive remediation when urgency allows. A serious incident should not be marked fully resolved because alert volume stopped; closure needs containment proof, recovery evidence and owners for longer-term vulnerabilities that remain.
For an identity compromise affecting a cloud application, put the sequence on one timeline: risky sign-in, token issuance, application activity, Azure role changes, affected device evidence, first alert, triage decision and containment action. Note the identity and clock source for every timestamp. If the responder revokes sessions but the attacker retains a valid application credential, the original trust boundary has not been restored. Likewise, isolating one host may leave an overprivileged managed identity attached to another exposed resource. Each containment choice needs an expected result and a check that the next malicious action would now fail.
Evidence protection should be proportional to the incident. Keep the original query, time range and relevant event identifiers; don’t reduce investigation to screenshots copied into a ticket. Verify that automation did not erase the only useful artifacts, and record who authorized disruptive containment. A post-incident review should close the specific credential, identity, network or software weakness that made the attack possible rather than just set the portal incident state to resolved.
Document uncertainty and handoff decisions during response
An incident can change owner several times as evidence moves from identity to endpoint, data or cloud infrastructure. Preserve unresolved hypotheses and explicit decisions rather than silently turning one analyst’s guess into accepted fact. A handoff note should identify current containment state, new evidence since the last update, sources that remain unavailable and the next authorized action. Avoid copying sensitive artifacts into broad channels merely to keep the summary convenient for every responder.
During a tabletop, have the first analyst hand over an incomplete case mid-investigation. Ask the second to reproduce one important query, explain which entities remain suspect and identify who is allowed to revoke the privileged account. Any missing information becomes a runbook improvement. Incident continuity is part of containment quality because delayed or contradictory decisions can let a real attacker keep access after the first alert was acknowledged.
Turn incident lessons into durable engineering changes
Post-incident review should improve prevention, detection and operational response. If a vulnerable public endpoint made compromise possible, revise the deployment policy. If excessive privilege enabled lateral movement, change the role pattern and review similar identities. If the SOC could not reconstruct activity because logs were missing, update collection and retention. Each action should be a bounded change with owner, review date and validation method. Counting closed investigations is a poor measure when the same root cause reappears every month.
Review one incident as an engineering product: identify the first detectable event, the action that contained the attacker, the unresolved coverage gaps and the safeguard that would have shortened the response. Convert that knowledge into a validated hunting query or detection, a configuration guardrail and a runbook improvement where justified. Rehearse the improved response with synthetic data. A unified Microsoft incident surface provides the context, but the value comes from teams actually changing their control system.