Lesson 68 — Correlating Microsoft Defender XDR and Sentinel Evidence
Follow an investigation across Microsoft Sentinel and Microsoft Defender XDR, connect alerts with endpoint activity, and build a defensible timeline without assuming every data source lives in the same query workspace.

What you will learn
Use alerts as leads, verify the underlying records and distinguish linked evidence from coincidence.
Agent Foskett case file — Two consoles, one suspicious device
A Sentinel alert flags unusual authentication activity involving a privileged account. Minutes later, a Defender XDR alert points to suspicious PowerShell execution on a workstation. Are these parts of one intrusion, two unrelated alerts or an authorized administrative operation? Agent Foskett must verify identities, devices, timestamps and underlying events before linking the stories.
Learning objectives
- Explain the different roles of Microsoft Sentinel and Microsoft Defender XDR in an investigation.
- Identify where Sentinel Logs KQL and Defender XDR advanced hunting queries run.
- Inspect Sentinel alerts and incidents without treating alert text as proof.
- Use endpoint network and process evidence to validate an alert hypothesis.
- Correlate records using stable identifiers, time windows and uncertainty notes.
- Document an evidence-led conclusion and appropriate escalation.
Why cross-platform correlation matters
Sentinel can aggregate signals from many sources, while Defender XDR provides detailed security telemetry from supported Microsoft workloads and endpoints. A single alert can be incomplete or misleading. Combining corroborating records can establish a more reliable sequence of events, provided collection and identity mappings are verified.
A hypothesis with alternatives
Hypothesis: an unusual privileged sign-in was followed by attacker-driven command execution and outbound network communication on the same device. Alternatives: an approved remote support session, a deployment script, benign endpoint detection or two events that merely occurred near one another.
Critical distinction — two KQL environments
Microsoft Sentinel Logs: query the Log Analytics workspace and tables actually ingested there, such as SecurityEvent, SecurityAlert and, where available, SecurityIncident. Microsoft Defender XDR advanced hunting: query Defender tables such as DeviceProcessEvents and DeviceNetworkEvents in the Defender portal. Their presence in Sentinel Logs depends on your integration and configuration. Do not paste a cross-product join into one portal unless all referenced tables are available in that query context.
Correlation keys and limitations
| Evidence | Useful correlation key | Common pitfall |
|---|---|---|
| Device | DeviceId, hostname, device inventory | Renames, short names, reused hostnames |
| Identity | UPN, SID, account identifiers | Different formats, shared admin accounts |
| Time | UTC timestamps and event sequence | Clock skew and ingestion delay |
| Process | Process ID plus creation time or unique ID | Process IDs can be reused |
| Alert | Alert ID and source record | Alert title alone is not raw evidence |
An incident groups related alerts for investigation, but grouping does not independently prove all activity belongs to one attack.
Verify telemetry before drawing conclusions
Confirm the relevant Windows event collection, Defender onboarding, advanced-hunting retention, Sentinel connector status and permissions. Check which tables exist in your workspace and whether the investigation period is retained. An empty result can mean missing telemetry, not absence of activity.
KQL hunt 1 — Verify Sentinel alert coverage
SecurityAlert
| where TimeGenerated >= ago(7d)
| summarize Alerts = count(),
LastSeen = max(TimeGenerated)
by ProductName, AlertName
| order by Alerts desc
| take 50Run in Sentinel Logs. The available alert fields and product names depend on your connected products and ingestion configuration. This is a coverage and prioritization check, not a detection verdict.
Start with the alert, not the assumption
Open the source alert and capture its alert identifier, detection time, affected entities, severity, detection logic and investigation link. Then locate the raw event or endpoint evidence that supports the detection. Preserve both event time and ingestion time when they differ.
KQL hunt 2 — Review recent Sentinel alert leads
SecurityAlert
| where TimeGenerated >= ago(48h)
| project TimeGenerated, AlertName, AlertSeverity,
ProductName, CompromisedEntity, SystemAlertId,
Description
| order by TimeGenerated desc
| take 200Run in Sentinel Logs. CompromisedEntity may be empty or may not match the endpoint hostname; inspect the alert entities in the incident or raw alert before using it as a join key. The row limit can omit alerts in busy workspaces.
Understand incident context
Where SecurityIncident is populated, review incident ownership, status, related alerts and investigation history. Incidents may be created or updated after the underlying events. Use the incident to organize the case; use original telemetry to substantiate each step.
KQL hunt 3 — Review incident updates
SecurityIncident
| where TimeGenerated >= ago(7d)
| summarize arg_max(TimeGenerated, *) by IncidentNumber
| project TimeGenerated, IncidentNumber, Title,
Severity, Status, Owner, CreatedTime, LastModifiedTime
| order by TimeGenerated desc
| take 100Run only if your Sentinel workspace has the SecurityIncident table. This query shows the latest ingested version per incident number; incident update timestamps are not necessarily the times of attacker activity.
Pivot to Defender XDR
In Microsoft Defender XDR, locate the affected device and review its timeline, related alerts and investigation package. Confirm device identity using the portal’s device identifier or asset inventory before comparing with Sentinel hostnames. Note the user context and process lineage.
KQL hunt 4 — Examine endpoint process execution
DeviceProcessEvents
| where Timestamp >= ago(48h)
| where DeviceName =~ "WORKSTATION-01"
| where FileName in~ ("powershell.exe", "pwsh.exe", "cmd.exe")
| project Timestamp, DeviceName, DeviceId,
FileName, ProcessCommandLine, AccountName,
InitiatingProcessFileName, InitiatingProcessCommandLine,
ProcessId
| order by Timestamp asc
| take 250Run in Defender XDR advanced hunting, not in Sentinel Logs unless this table is separately available there. Replace WORKSTATION-01 with a verified lab device. Command interpreters are common administrative tools; validate execution context and parent process before escalation.
Process context beats filename alone
A PowerShell process can be expected during administration or suspicious during an intrusion. Inspect the parent process, full command line, signing information, account, device and change window. When available, use unique process identifiers and process creation time rather than assuming a numeric PID identifies one process forever.
KQL hunt 5 — Examine endpoint outbound connections
DeviceNetworkEvents
| where Timestamp >= ago(48h)
| where DeviceName =~ "WORKSTATION-01"
| project Timestamp, DeviceName, DeviceId,
RemoteIP, RemoteUrl, RemotePort,
InitiatingProcessFileName,
InitiatingProcessCommandLine,
InitiatingProcessAccountName
| order by Timestamp asc
| take 250Run in Defender XDR advanced hunting. Compare destinations and initiating processes with approved software, DNS evidence and other device activity. A remote connection alone does not establish command-and-control activity.
Build a normalized timeline
Export or document the relevant records from each product using approved investigation procedures. Normalize timestamps to UTC and record source, event ID or alert ID, device identifier, account, observed action and confidence. Correlate on multiple dimensions; proximity in time is not enough.
A worked correlation example
| Illustrative UTC time | Source | Observation | Validation needed |
|---|---|---|---|
| 01:12 | Sentinel / sign-in telemetry | Unusual privileged sign-in alert | Original sign-in and account details |
| 01:17 | Defender XDR process telemetry | PowerShell on investigated device | Process lineage and user context |
| 01:19 | Defender XDR network telemetry | Outbound connection from same device | Destination ownership and initiating process |
| 01:25 | Change management | Scheduled support activity recorded | Approved scope and actual operator |
These are fictional training observations, not real query results. The timeline provides a hypothesis to test, not a confirmed attack.
Common correlation mistakes
- Joining devices by hostname without verifying renames or duplicates.
- Treating ingestion time as the original activity time.
- Assuming an incident title proves execution or compromise.
- Assuming Defender advanced-hunting tables are automatically present in Sentinel Logs.
- Linking events based only on a narrow time window.
- Ignoring legitimate change and endpoint management activity.
MITRE ATT&CK mapping
Potential hypotheses include Valid Accounts (T1078), Command and Scripting Interpreter: PowerShell (T1059.001) and Application Layer Protocol (T1071). Map techniques only where the underlying behavior is observed and supported; an alert name or tool choice does not establish attacker intent.
Agent Foskett investigation workflow
When to escalate
Escalate under the organization’s incident response process when unauthorized access, suspicious process behavior and relevant network activity are corroborated. Preserve evidence before containment actions and follow approved procedures; do not isolate production devices based solely on coincident alerts.
Case file resolution — Two consoles, one suspicious device
The unusual sign-in and PowerShell execution were both associated with an approved remote support session. The endpoint process lineage matched the signed management agent, the outbound destination belonged to the support platform, and the change record named the operator and time window. Agent Foskett records the linked evidence and closes the malicious-activity hypothesis as unsupported.
What the investigation still revealed
The support team had not documented the workstation naming change, which initially made cross-platform device matching uncertain. The analyst records this as a visibility and asset-inventory improvement, separate from any claim of compromise.
Hands-on lab — Correlate Sentinel and Defender evidence
- Confirm Sentinel
SecurityAlertand, if applicable,SecurityIncidentcoverage. - Select an authorized test incident or benign training alert.
- Capture source alert ID, event time, affected account and device details.
- Verify the corresponding device identity in Defender XDR.
- Run the process and network hunts against a permitted lab device.
- Create a UTC timeline and annotate at least two alternative explanations.
- Write a conclusion distinguishing observed facts from inference.
Knowledge check
- Why can’t you assume
DeviceProcessEventsexists in Sentinel Logs? - What makes a device identifier more reliable than a hostname alone?
- Why is a Sentinel incident not proof of compromise?
- How can ingestion delay distort a timeline?
- What evidence would link a network connection to a suspicious process?
Operational quality checklist
- Confirm connector, table and retention coverage.
- Run each query in its stated environment.
- Normalize timestamps and verify account/device identifiers.
- Keep exploratory queries time-bounded and read-only.
- Preserve raw evidence references and alternative explanations.
- Do not publish links to unfinished Lesson 69.
Agent Foskett takeaway
One alert starts the investigation. Correlated, validated evidence determines the conclusion. Treat every cross-platform match as a claim to verify, not a shortcut to certainty.
The Logs Already Knew.
