Lesson 69 — Mapping Advanced Threat Hunts to MITRE ATT&CK
Translate suspicious events into evidence-based MITRE ATT&CK mappings, connect identity, endpoint and network observations, and explain what the logs do—and do not—prove.

What you will learn
Move from raw telemetry to a well-supported technique hypothesis without treating a single event as proof of compromise.
Agent Foskett case file — Five clues, one hypothesis
An analyst notices an unusual sign-in, a privileged group change and repeated outbound traffic from the same workstation. A dashboard suggests an attack chain. Agent Foskett must determine which behaviours are actually supported by the collected logs, which ATT&CK techniques fit, and which links remain unproven.
Learning objectives
- Describe the difference between ATT&CK tactics, techniques and sub-techniques.
- Identify observable behaviours in Windows, Entra and network telemetry.
- Use KQL to collect and validate candidate evidence.
- Map supported observations to ATT&CK without inferring intent from an event ID alone.
- Build an investigation matrix with timestamps, source, confidence and gaps.
- Present a cautious conclusion and recommend next investigative steps.
What MITRE ATT&CK represents
MITRE ATT&CK is a knowledge base of adversary tactics and techniques derived from real-world observations. A tactic describes the adversary objective (the “why”); a technique describes a way to achieve it (the “how”). A sub-technique adds specificity. ATT&CK is not a verdict engine, and an ATT&CK label is not proof that an event was malicious.
Map behaviour before assigning a label
Observation: a security-enabled local group gained a member. Possible mapping: Account Manipulation (T1098), if the broader evidence supports an adversary altering an account or its privileges. Alternative: an approved access request. Record the raw event and administrative context before assigning the technique.
Telemetry boundaries — know your workspace
These exercises run in Microsoft Sentinel Logs and use SecurityEvent, SigninLogs, AuditLogs, CommonSecurityLog and SecurityAlert. Each table requires its corresponding connector and collection settings. Defender XDR advanced-hunting tables such as DeviceProcessEvents are not assumed to exist in this workspace. Query a table only when available, and validate its actual schema.
ATT&CK mapping quick reference
| Observed behaviour | Candidate technique | Evidence needed |
|---|---|---|
| Unexpected account privilege change | T1098 — Account Manipulation | Who changed what, approval, resulting permissions |
| New scheduled task | T1053.005 — Scheduled Task | Task action, creator, execution and purpose |
| Remote service or host access | T1021 — Remote Services (appropriate sub-technique) | Protocol, source/destination, authentication, execution |
| Regular suspicious outbound connections | T1071 — Application Layer Protocol (when supported) | Protocol, process, destination, timing and content/context |
| Suspicious PowerShell execution | T1059.001 — PowerShell | Process/script evidence, arguments, context |
These are candidate mappings. Choose a specific sub-technique only when evidence supports its details.
Tactic versus technique
A technique can contribute to more than one adversary objective. Do not automatically infer the tactic from the tool name, event type or the stage of an imagined kill chain. For example, a remote logon may be normal administration; remote services become an ATT&CK hunting lead only when the observed behaviour and context support that hypothesis.
Start with collection coverage
Before hunting, check the time range, host coverage, connector health and audit policy. Missing telemetry weakens confidence; it does not prove the behaviour did not happen.
KQL hunt 1 — Check relevant Windows event coverage
SecurityEvent
| where TimeGenerated >= ago(7d)
| where EventID in (4624, 4672, 4698, 4720, 4728, 4732, 4756)
| summarize Events = count(), Hosts = dcount(Computer),
LastSeen = max(TimeGenerated) by EventID
| order by EventID ascThe result describes collected events only. Event 4672 identifies special privileges assigned to a logon; by itself it does not establish privilege escalation.
Privilege-change evidence
Group membership changes (for example, event 4732 for a security-enabled local group) can support a privilege-modification hypothesis. Verify the target group, actor, member and approved change. Events 4728 and 4756 cover other security group scopes; their interpretation depends on the group type and reporting host.
KQL hunt 2 — Triage group membership changes
SecurityEvent
| where TimeGenerated >= ago(7d)
| where EventID in (4728, 4732, 4756)
| project TimeGenerated, Computer, EventID,
Account, Activity, EventData
| order by TimeGenerated desc
| take 200Inspect raw EventData for the actor, target group and member. The result cap is for triage and can omit activity in busy environments.
Identity activity is context, not attribution
An unusual Microsoft Entra sign-in can provide context for a later privilege change, but matching usernames or nearby timestamps alone cannot prove one caused the other. Compare request identifiers, device, IP address, session information and confirmed account ownership where available.
KQL hunt 3 — Review Entra sign-in outcomes
SigninLogs
| where TimeGenerated >= ago(7d)
| summarize Attempts = count(),
Failures = countif(tostring(ResultType) != "0"),
LastSeen = max(TimeGenerated)
by UserPrincipalName, IPAddress
| where Attempts >= 5 or Failures > 0
| order by Failures desc, Attempts desc
| take 100ResultType 0 commonly indicates success. Review the actual result codes and Conditional Access context before interpreting a failed or unusual sign-in.
AuditLogs and role changes
Microsoft Entra role assignments are represented in AuditLogs when relevant auditing is ingested. The TargetResources field is dynamic; inspect it before assuming the array position or the shape of modified properties. Correlate the operation with change requests and the initiator.
KQL hunt 4 — Review directory role activity
AuditLogs
| where TimeGenerated >= ago(14d)
| where Category =~ "RoleManagement"
| project TimeGenerated, OperationName, Result,
InitiatedBy, TargetResources, CorrelationId
| order by TimeGenerated desc
| take 150Operation names and payloads vary. A role-management operation is a lead for review, not an automatic indicator of attacker privilege escalation.
Network patterns and ATT&CK
Periodic outbound communication may suggest beaconing, but legitimate update services and monitoring agents can produce similar intervals. A mapping to an application-layer command-and-control technique requires protocol and behavioural evidence, ideally corroborated by endpoint process context.
KQL hunt 5 — Summarise outbound destinations
CommonSecurityLog
| where TimeGenerated >= ago(24h)
| where isnotempty(SourceIP) and isnotempty(DestinationIP)
| summarize Connections = count(),
FirstSeen = min(TimeGenerated),
LastSeen = max(TimeGenerated)
by SourceIP, DestinationIP, DeviceAction
| where Connections >= 10
| order by Connections desc
| take 100This is a frequency summary, not a beaconing detector. CommonSecurityLog fields and action values depend on the reporting device and parser; confirm ingestion and filtering semantics.
When to assign a technique
- State the behaviour in plain language before selecting an ATT&CK ID.
- Link the technique to the original event or query output.
- Explain why the observed behaviour fits the technique definition.
- Record legitimate alternatives and what would disprove the hypothesis.
- Avoid sub-technique precision unsupported by the evidence.
Build an evidence-to-technique matrix
| Time / source | Observation | Candidate ATT&CK | Confidence / gap |
|---|---|---|---|
| 09:14 • SigninLogs | Unusual sign-in from unfamiliar IP | Not mapped yet | Needs identity and session verification |
| 09:21 • SecurityEvent | Member added to local privileged group | T1098 (candidate) | Actor and approval not yet verified |
| 09:35 • Network log | Repeated outbound traffic | Not mapped yet | Needs protocol/process evidence |
The entries above are illustrative case-file data, not live telemetry or confirmed malicious events.
Confidence and evidence quality
Use a documented confidence rubric rather than a made-up percentage. For example, low means a weak or ambiguous indicator; moderate means multiple consistent observations with important gaps; high means independently corroborated behaviour with competing explanations evaluated. Confidence in a technique mapping is distinct from confidence that an incident is malicious.
Avoid ATT&CK over-mapping
One suspicious logon should not automatically be mapped to Initial Access, Credential Access, Lateral Movement and Persistence. Multiple tactics require separate supporting observations. A detailed-looking matrix with unsupported labels is less useful than a short, accurate one.
Agent Foskett investigation workflow
Cross-platform correlation limits
Use consistent UTC timestamps and normalise host/account identifiers carefully. Microsoft Sentinel incidents, Windows security events and Defender XDR advanced hunting can use different schemas and retention periods. Do not write a cross-product join until the required tables and permissions are confirmed.
Case file resolution — Five clues, one hypothesis
The privileged group change is traced to an approved emergency access request. The sign-in belongs to the on-call administrator, while the outbound traffic is generated by an approved monitoring agent. Agent Foskett records the candidate ATT&CK mapping for the group-change behaviour as an investigative hypothesis, but does not claim an attacker performed it. The apparent attack chain is unsupported by the corroborating evidence.
Finding: several suspicious-looking observations, no confirmed adversary technique or malicious sequence.
Hands-on lab — Build an ATT&CK evidence map
- Confirm that the relevant Sentinel tables are populated.
- Run the Windows event coverage query and record gaps.
- Review a group membership change and inspect its raw payload.
- Review Entra sign-in and role-management context where available.
- Summarise outbound network destinations if firewall logs are ingested.
- Create a matrix with observation, source, timestamp, candidate technique and confidence.
- List at least two benign alternatives and the evidence needed to distinguish them.
- Write a final conclusion that separates observed facts from hypotheses.
Knowledge check
- What is the difference between an ATT&CK tactic and technique?
- Why does event 4672 not prove privilege escalation?
- When is it appropriate to assign an ATT&CK sub-technique?
- What evidence would strengthen a suspected C2 mapping?
- Why should confidence in a technique mapping differ from confidence in malicious intent?
Operational quality checklist
- Correct table, connector and event semantics confirmed.
- Original records preserved with timestamps and source.
- No ATT&CK technique assigned solely from an alert name.
- Benign alternatives considered.
- Confidence and missing telemetry documented.
- KQL queries remain read-only and time bounded.
- No links to unpublished Lesson 70.
Agent Foskett takeaway
ATT&CK is a map of behaviours, not a shortcut to attribution. Start with the logs, test the hypothesis and map only what the evidence can support.
The Logs Already Knew.
