Skip to content

Azure Incident Investigation

A strong Azure incident investigation is not just about closing an alert. It is about determining what happened, how far the activity spread, what evidence supports the conclusion, how to contain the threat safely, and how to prevent recurrence.

Welcome to the Azure Incident Investigation Runbook.

This runbook brings together the practical skills from:

  • Microsoft Sentinel Lab
  • Microsoft Defender for Cloud Lab
  • Microsoft Entra ID Security Lab
  • Azure Network Security Lab
  • Azure Security Assessment Runbook
  • Microsoft Entra ID Security Review Runbook
  • Azure Network Security Review Runbook

The objective is to move from:

I can investigate an individual Azure alert.

to:

I can perform a structured cloud incident investigation from initial detection through containment, recovery, and post-incident reporting.

This runbook is relevant to roles such as:

  • Security Operations Analyst
  • Cloud Security Analyst
  • Azure Security Engineer
  • Incident Responder
  • Microsoft Sentinel Analyst
  • Cloud Security Consultant
  • Security Engineer
  • Cloud Security Architect
Item Details
Runbook Azure Incident Investigation
Difficulty Intermediate β†’ Advanced
Estimated Time Depends on incident scope
Primary Focus Detection, Investigation, Containment & Recovery
Target Roles SOC Analyst, Azure Security Engineer, Incident Responder
Certification Alignment SC-200, AZ-500, SC-300, SC-100
Output Incident Timeline, Evidence Register, Scope Assessment, Root Cause, Incident Report

You are part of the security team responsible for an Azure environment.

Microsoft Sentinel reports suspicious activity involving a privileged identity.

Shortly afterward:

  • an NSG is modified
  • a new application credential appears
  • Defender for Cloud reports suspicious activity on a VM
  • unusual outbound network traffic is observed

Management asks:

β€œHas the Azure environment been compromised, what has been affected, and what should we do now?”

Your job is to determine:

  • whether the alert is legitimate
  • how the activity began
  • which identities are affected
  • which resources were changed
  • whether persistence exists
  • whether data or workloads were accessed
  • how far the attacker may have moved
  • how to contain the incident
  • how to restore secure operations

By the end of this runbook, you should be able to:

  • open and classify an incident
  • validate Sentinel alerts
  • investigate Microsoft Entra sign-ins
  • review Entra audit activity
  • investigate Azure Activity Logs
  • correlate Defender for Cloud findings
  • investigate suspicious RBAC changes
  • investigate compromised identities
  • investigate application identities
  • investigate compromised VMs
  • review network evidence
  • build an incident timeline
  • determine incident scope
  • identify persistence
  • preserve evidence
  • contain affected identities and workloads
  • support eradication and recovery
  • determine root cause
  • document lessons learned

Before performing incident-response actions, confirm:

  • tenant and subscriptions in scope
  • incident-response authority
  • whether accounts may be disabled
  • whether sessions may be revoked
  • whether resources may be isolated
  • whether VMs may be stopped
  • whether snapshots may be collected
  • whether network controls may be modified
  • evidence-retention requirements
  • legal/compliance escalation requirements

Do not jump immediately from:

Alert

to:

Delete everything.

Containment should reduce risk while preserving your ability to understand what happened.

Use:

Prepare
↓
Detect
↓
Validate
↓
Classify
↓
Scope
↓
Contain
↓
Preserve Evidence
↓
Investigate
↓
Eradicate
↓
Recover
↓
Root Cause
↓
Lessons Learned

Assign a unique incident ID.

Example:

AZ-IR-2026-001

Record:

Field Details
Incident ID
Detection Time
Detection Source
Initial Severity
Tenant
Subscription
Incident Owner
Status

Record the original detection exactly as received.

Possible sources:

  • Microsoft Sentinel

  • Microsoft Defender for Cloud

  • Microsoft Defender

  • Microsoft Entra ID

  • Azure Monitor

  • user report

  • external intelligence

Create:

Alert Source:
Alert Name:
Severity:
Affected Identity:
Affected Resource:
Time:
Description:
Evidence:

Do not assume an alert means confirmed compromise.

Determine whether the activity is:

False Positive
Benign Positive
Suspicious
Confirmed Malicious

Ask:

Was the user expected to perform this activity?

Was there an approved change?

Is the location unusual because of VPN or travel?

Is automation responsible?

Does another telemetry source confirm suspicious behavior?

Use a consistent model.

Examples:

  • confirmed Global Administrator compromise

  • broad tenant takeover

  • major sensitive-data exposure

  • multiple subscriptions affected

Examples:

  • privileged account compromise

  • production workload compromise

  • unauthorized security-control changes

Examples:

  • suspicious activity with limited scope

  • lower-privilege compromised identity

Examples:

  • minor suspicious activity

  • likely benign event requiring confirmation

Start with:

First Known Suspicious Event
↓
Look Backward
↓
Look Forward

Search backward for:

  • initial access

  • credential compromise

  • suspicious authentication

  • reconnaissance

Search forward for:

  • privilege escalation

  • resource changes

  • persistence

  • lateral movement

  • data access

Determine whether the activity involves:

  • Entra user

  • privileged administrator

  • guest

  • service principal

  • managed identity

  • automation identity

Record:

Identity Type Privileged? Expected Activity?

Ask:

Which identity actually performed the action?

Do not assume the visible human account was the only identity involved.

For the affected user review:

  • timestamp

  • source IP

  • location

  • application

  • authentication result

  • MFA

  • device

  • Conditional Access result

Create:

Time IP Location Application Result MFA Risk

A successful sign-in proves:

authentication succeeded.

It does not prove:

the user was legitimate.

08 β€” Investigate Suspicious Authentication Patterns

Section titled β€œ08 β€” Investigate Suspicious Authentication Patterns”

Look for:

  • unusual geography

  • unfamiliar IP

  • unfamiliar device

  • repeated failures followed by success

  • unexpected privileged-account use

  • unusual applications

20 Failed Sign-Ins
↓
Successful Authentication
↓
Privileged Azure Change

This should receive immediate attention.

Determine:

  • which policies applied

  • whether MFA was required

  • whether policies were bypassed

  • whether the identity was excluded

An unexplained Conditional Access exclusion may itself be part of the incident.

Search for high-impact identity events.

Examples:

  • role assignments

  • group membership changes

  • application registration changes

  • credential creation

  • Conditional Access changes

  • guest creation

Suspicious Sign-In
↓
Privilege Change
↓
Persistence

If a user suddenly receives a privileged role:

ask:

  • who assigned it?

  • when?

  • was PIM involved?

  • was the change approved?

  • what happened afterward?

Time Actor Target Role Legitimate?

Where PIM is used, investigate:

  • role activation

  • approver

  • justification

  • activation duration

  • MFA

  • unusual timing

A legitimate PIM activation has a very different context from an unauthorized permanent role assignment.

Look for suspicious changes to:

  • app registrations

  • service principals

  • permissions

  • client secrets

  • certificates

Compromised Administrator
↓
Creates App Credential
↓
Service Principal
↓
Persistent Access

This can remain even after the original user credential is reset.

Review:

  • application

  • credential creation time

  • actor

  • expiry

  • permissions

Ask:

Did the attacker create another way back into the environment?

15 β€” Search for New or Modified Service Principals

Section titled β€œ15 β€” Search for New or Modified Service Principals”

Investigate:

  • unexpected application identities

  • new admin consent

  • permission increases

  • new credentials

  • ownership changes

If broad application permissions were granted:

determine:

  • who approved the consent

  • which application

  • which permissions

  • whether business justification exists

Investigate changes involving:

  • Owner

  • Contributor

  • User Access Administrator

  • custom roles

  • role assignments

An attacker may move from:

Entra Privilege
↓
Azure RBAC
↓
Subscription Access

Azure Activity helps answer:

Who changed Azure resources?

Search for:

  • VM operations

  • NSG changes

  • route-table changes

  • public IP creation

  • resource deployment

  • Key Vault changes

  • storage changes

  • RBAC changes

Time Identity Operation Resource Result

Look for changes involving:

  • NSGs

  • route tables

  • firewall rules

  • public IPs

  • peerings

Compromised Identity
↓
NSG Changed
↓
Public Port Opened
↓
Workload Accessed

For related alerts document:

  • affected resource

  • severity

  • time

  • alert type

  • evidence

  • recommended response

Defender tells you:

suspicious workload behavior may exist.

Azure Activity may tell you:

who changed the workload.

Entra may tell you:

how that identity authenticated.

Combine them.

Open the Sentinel incident.

Review:

  • alert sources

  • entities

  • timeline

  • related incidents

  • severity

  • owner

Possible entities:

  • account

  • IP

  • host

  • Azure resource

  • URL

  • file

Create an entity map.

User A
↓
IP X
↓
Azure VM
↓
NSG Change
↓
External IP Z

Create one chronological timeline combining:

  • Entra sign-ins

  • Entra audit

  • PIM

  • Azure Activity

  • Defender alerts

  • Sentinel incidents

  • network telemetry

Time Event Source Identity/Resource Significance

The incident should eventually become a story you can explain clearly.

Record:

  • VM name

  • subscription

  • resource group

  • public IP

  • private IP

  • subnet

  • NSG

  • managed identity

  • disks

  • Defender status

Ask:

  • does it have a public IP?

  • are SSH/RDP ports exposed?

  • which NSGs apply?

  • was exposure recently changed?

Determine whether the VM uses:

  • system-assigned managed identity

  • user-assigned managed identity

  • service principal credentials

Could the compromised workload access Azure resources through its identity?

Search Azure activity and relevant logs for actions performed using the workload identity.

Determine:

  • resources accessed

  • permissions

  • timing

  • unusual operations

Use available telemetry such as:

  • Azure Firewall

  • NSG-related flow information where available

  • Sentinel network logs

  • Defender context

Investigate:

  • source

  • destination

  • port

  • protocol

  • time

  • volume

Ask:

Where did this workload communicate before and after the alert?

Example:

Azure VM
↓
Unexpected External IP
↓
Repeated Outbound Connections

Ask:

  • legitimate destination?

  • application dependency?

  • threat intelligence?

  • other compromised workloads?

Review communication between:

  • VM to VM

  • frontend to application

  • application to database

  • development to production

  • spoke to spoke

May occur through:

Network Connectivity
+
Azure RBAC
+
Managed Identity
+
Application Permissions

Do not investigate only IP traffic.

If storage is involved, review:

  • network configuration

  • role assignments

  • access mechanisms

  • recent configuration changes

  • data access evidence where available

Ask:

Was sensitive data accessed, modified, deleted, or exposed?

Investigate:

  • secret access

  • key access

  • permission changes

  • public access changes

  • credential updates

An attacker may retrieve:

  • application secrets

  • certificates

  • keys

which could expand the incident.

Possible initial-access paths may include:

  • stolen password

  • stolen session

  • weak MFA

  • exposed application secret

  • compromised VM

  • exposed administrative port

  • over-trusted application

Do not write:

Sentinel alert

as the root cause.

Sentinel detected the behavior.

Look for:

  • Entra role assignment

  • Azure RBAC escalation

  • PIM misuse

  • group membership changes

  • application permissions

Look for:

  • new users

  • new service principals

  • new app secrets

  • new certificates

  • new privileged group memberships

  • new RBAC assignments

  • automation resources

Ask:

If the original identity is disabled, can the attacker still return?

Review whether the attacker changed:

  • Conditional Access

  • logging

  • Sentinel configuration

  • Defender settings

  • NSGs

  • diagnostic settings

Security-control modification can indicate deliberate defense evasion.

Do not assume the incident is limited to one subscription.

Review cross-subscription access where the identity has permissions.

For privileged Entra compromise, review:

  • applications

  • directory roles

  • guest access

  • service principals

  • security-policy changes

Category Affected? Evidence
Entra Users
Privileged Roles
Service Principals
Azure RBAC
VMs
Storage
Key Vault
Networking
Subscriptions

Use:

Identity Permissions
+
Network Reachability
+
Application Permissions
+
Data Access
=
Potential Blast Radius

Depending on authorization and investigation needs, preserve:

  • Sentinel incident data

  • Entra sign-in logs

  • Entra audit logs

  • Azure Activity

  • Defender alerts

  • network telemetry

  • relevant VM/disk evidence

Capture important evidence before actions that may change or destroy it.

Create:

Evidence ID Source Time Collected Description Owner

Where required and authorized, consider appropriate Azure disk/snapshot evidence methods before destructive remediation.

Document:

  • resource

  • time

  • snapshot/reference

  • investigator

Depending on incident severity and authority, containment may include:

  • disable user

  • revoke sessions

  • reset credentials

  • remove privileged roles

  • remove suspicious group membership

Before disabling a critical service identity or administrator:

understand business impact.

Potential actions may include:

  • remove suspicious secret/certificate

  • disable or restrict the application identity

  • reduce permissions

  • revoke unnecessary consent

Possible actions may include:

  • isolate with controlled NSG

  • remove public exposure

  • restrict outbound traffic

  • preserve required forensic access

Conceptually:

Compromised VM
↓
Quarantine Network Control
↓
Restricted Communication

Potentially restrict:

  • public administrative access

  • malicious destination IPs

  • compromised workload communication

  • unnecessary peer routes

Only where authorized.

Contain enough to stop further damage.

Do not destroy evidence unnecessarily.

After evidence is preserved and unauthorized access is confirmed:

remove:

  • malicious directory roles

  • suspicious Azure RBAC assignments

  • malicious group memberships

  • unauthorized PIM assignments

Remove verified malicious:

  • application credentials

  • service principals

  • users

  • role assignments

  • automation resources

Do not delete unfamiliar objects simply because you do not recognize them.

Validate business ownership first.

Depending on scope, rotate:

  • user credentials

  • application secrets

  • certificates

  • Key Vault secrets

  • workload credentials

For a confirmed VM compromise, determine whether:

  • patching

  • cleaning

  • rebuilding

provides sufficient assurance.

Often a rebuild from a known-good image may be preferable to trusting a heavily compromised system.

Ensure that attackers did not leave weakened controls such as:

  • open NSGs

  • broad RBAC

  • Conditional Access exclusions

  • disabled logging

  • public endpoints

Recovery should include:

  • restoring required application functionality

  • validating identity

  • validating networking

  • validating access

  • validating logging

  • validating Defender/Sentinel

During the agreed observation period, monitor:

  • affected users

  • application identities

  • IP addresses

  • Azure resources

  • security alerts

  • network activity

Document:

Root Cause:
Initial Access:
Contributing Factors:
Control Failure:
Detection Source:

Do not write:

Root Cause: Microsoft Sentinel alert.

Write:

Root Cause: A long-lived application credential was exposed and used to obtain unauthorized access.

Examples:

  • excessive privilege

  • missing MFA

  • Conditional Access exception

  • public RDP

  • over-privileged service principal

  • incomplete logging

  • weak segmentation

Consider:

  • confidentiality

  • integrity

  • availability

  • regulatory impact

  • operational impact

  • customer impact

Instead of:

Microsoft.Authorization/roleAssignments/write occurred.

Explain:

An unauthorized identity granted additional privileged access within the Azure subscription, increasing the attacker’s control over cloud resources.

Use:

Initial Access
↓
Authentication
↓
Privilege Escalation
↓
Persistence
↓
Resource Changes
↓
Detection
↓
Containment
↓
Eradication
↓
Recovery

Create:

Control Gap Corrective Action Owner Priority

Examples:

  • weak privileged authentication

  • long-lived secrets

  • excessive Owner access

  • missing Sentinel detection

  • public management access

Ask:

What could have prevented the incident?

Why was it detected when it was?

What slowed containment?

Was telemetry sufficient?

What made restoration difficult?

Potential improvements may include:

  • stronger Conditional Access

  • PIM

  • managed identities

  • reduced Azure RBAC

  • Private Endpoints

  • centralized network controls

  • Defender for Cloud

  • Sentinel detections

Convert lessons learned into new or improved detection.

Example:

Incident Pattern
↓
KQL Hunt
↓
Detection Logic
↓
Analytics Rule

Potential low-risk automation:

  • assign incident

  • tag incident

  • enrich IP

  • notify SOC

  • collect evidence

Higher-risk automation may include:

  • disable account

  • isolate workload

and should be carefully governed.

Recommended structure:

1. Executive Summary
2. Incident Overview
3. Severity
4. Detection
5. Scope
6. Timeline
7. Evidence
8. Identity Investigation
9. Azure Resource Investigation
10. Network Investigation
11. Containment
12. Root Cause
13. Business Impact
14. Eradication
15. Recovery
16. Corrective Actions
17. Lessons Learned
Incident ID:
AZ-IR-001
Severity:
High
Summary:
Suspicious authentication involving a privileged Microsoft Entra identity was followed by unauthorized Azure network and application-identity changes.
Affected Areas:
Microsoft Entra ID, Azure RBAC, network controls, and application identities.
Detection:
Microsoft Sentinel and Microsoft Defender for Cloud.
Containment:
The affected identity was restricted, suspicious privilege was removed, and impacted resources were isolated.
Root Cause:
Compromise of a privileged credential combined with excessive standing access.
Corrective Actions:
Strengthen privileged authentication, reduce standing access through PIM, improve application-identity governance, and expand Sentinel detection coverage.
ID Finding Severity Corrective Action
IR-001
IR-002

66 β€” Example Finding β€” Excessive Standing Privilege

Section titled β€œ66 β€” Example Finding β€” Excessive Standing Privilege”
Finding ID:
IR-001
Title:
Compromised Identity Held Permanent Privileged Access
Severity:
High
Observation:
The compromised human identity permanently held a high-impact administrative role.
Risk:
Credential compromise provided immediate privileged capability without additional elevation controls.
Recommendation:
Reduce standing privilege and implement controlled temporary privileged access through PIM where appropriate.
Finding ID:
IR-002
Title:
Application Credential Creation Was Not Detected
Severity:
High
Observation:
A new privileged application credential was created during the incident without generating a dedicated SOC alert.
Risk:
Attackers may establish persistent non-human access without timely detection.
Recommendation:
Implement dedicated Sentinel analytics for high-risk application-identity changes.

🎀 Azure Incident Investigation Interview Questions

Section titled β€œπŸŽ€ Azure Incident Investigation Interview Questions”

Practise without notes.

1. How would you investigate an Azure security incident?

Section titled β€œ1. How would you investigate an Azure security incident?”

3. What is the difference between validation and containment?

Section titled β€œ3. What is the difference between validation and containment?”

5. How would you investigate a suspicious Entra sign-in?

Section titled β€œ5. How would you investigate a suspicious Entra sign-in?”

7. How would you detect persistence through service principals?

Section titled β€œ7. How would you detect persistence through service principals?”

10. How would you investigate an unexpected NSG change?

Section titled β€œ10. How would you investigate an unexpected NSG change?”

12. How does Sentinel support Azure incident response?

Section titled β€œ12. How does Sentinel support Azure incident response?”

15. How would you use Defender for Cloud during an investigation?

Section titled β€œ15. How would you use Defender for Cloud during an investigation?”

16. How would you correlate Defender with identity activity?

Section titled β€œ16. How would you correlate Defender with identity activity?”

17. How would you investigate a potentially compromised Azure VM?

Section titled β€œ17. How would you investigate a potentially compromised Azure VM?”

21. How would you investigate suspicious outbound traffic?

Section titled β€œ21. How would you investigate suspicious outbound traffic?”

A Global Administrator signs in from an unusual location and creates a new client secret.

Use:

Validate Sign-In
↓
Authentication Context
↓
Audit Activity
↓
Application
↓
Permission
↓
Persistence
↓
Containment

Microsoft Defender reports suspicious activity on a production VM.

A strong answer should discuss:

  • validate Defender finding

  • identify VM

  • review public exposure

  • inspect identity

  • correlate Sentinel

  • review Azure Activity

  • review network evidence

  • isolate if required

  • determine scope

A compromised user is disabled. Is the incident closed?

No.

Check whether the attacker created:

  • service principals

  • secrets

  • roles

  • group memberships

  • Azure RBAC assignments

  • resources

An NSG was modified to expose RDP to the internet.

Investigate:

  • actor

  • timestamp

  • authentication context

  • affected workloads

  • subsequent connection attempts

  • whether the change was authorized

The Sentinel incident only contains one alert. How do you determine whether the incident is larger?

Expand beyond the incident itself.

Use:

  • Entra logs

  • audit logs

  • Azure Activity

  • Defender

  • network telemetry

  • application identities

Use:

Detect
↓
Validate
↓
Scope
↓
Preserve
↓
Contain
↓
Investigate
↓
Eradicate
↓
Recover
↓
Improve

Avoid:

β€œI would disable the account and close the incident.”

A stronger answer is:

β€œI would validate the alert, identify the affected identity and resources, preserve relevant evidence, determine the scope through Entra, Azure Activity, Sentinel, Defender, and network telemetry, contain the immediate threat, search for persistence, remediate the root cause, recover securely, and document corrective actions.”

That sounds like a Cloud Incident Responder.

Create sanitized examples of:

  • incident intake form

  • Sentinel investigation

  • identity timeline

  • Azure Activity timeline

  • affected-resource matrix

  • evidence register

  • incident attack-path diagram

  • containment plan

  • root-cause analysis

  • executive incident report

Instead of:

Azure incident response experience.

Use:

Investigated simulated Azure security incidents by correlating Microsoft Entra sign-ins, Entra audit activity, Azure Activity Logs, Microsoft Sentinel, Defender for Cloud, and network telemetry.

Or:

Performed end-to-end Azure incident-response exercises covering validation, scope analysis, compromised identities, suspicious Azure resource changes, persistence, containment, root-cause analysis, and recovery.

You should be able to:

  • validate Azure security alerts

  • investigate Entra sign-ins

  • review audit activity

  • investigate RBAC changes

  • review Azure Activity Logs

  • analyze Sentinel incidents

  • correlate Defender findings

  • investigate compromised VMs

  • review managed-identity activity

  • investigate network evidence

  • identify persistence

  • determine blast radius

  • preserve evidence

  • contain identities and workloads

  • determine root cause

  • create professional incident reports

The objective is not:

I closed the Sentinel incident.

The objective is:

I established what happened, which identities and resources were affected, how far the attacker may have moved, whether persistence exists, and what controls must change to prevent recurrence.

You now have a repeatable Azure incident-investigation methodology:

Incident Intake
↓
Alert Validation
↓
Identity Investigation
↓
Azure Resource Investigation
↓
Network Investigation
↓
Scope Analysis
↓
Containment
↓
Evidence Preservation
↓
Eradication
↓
Recovery
↓
Root Cause
↓
Lessons Learned

This directly supports roles such as:

  • Security Operations Analyst

  • Azure Security Engineer

  • Cloud Incident Responder

  • Cloud Security Engineer

  • Microsoft Sentinel Analyst

  • Security Consultant

You have now completed the Azure security lab and runbook sequence:

Azure RBAC Lab
↓
Microsoft Entra ID Security Lab
↓
Azure Network Security Lab
↓
Microsoft Defender for Cloud Lab
↓
Microsoft Sentinel Lab
↓
Azure Security Assessment Runbook
↓
Microsoft Entra ID Security Review Runbook
↓
Azure Network Security Review Runbook
↓
Azure Incident Investigation Runbook

At this point, the Azure section forms the same complete career-readiness loop as AWS:

Certification knowledge β†’ Labs β†’ Professional runbooks β†’ Job readiness β†’ Interview readiness.

➑️ Next: Google Cloud Certification & Security Path