READ-ONLY HUNT COMPANION / OCTOBER 19, 2026
Make the result explain itself.
A focused investigation aid for service-principal authentication and directory changes. It is not a tenant-clearance test.
Download the KQL Download the worksheet
Choose the query family for your platform
The full detection and response companion includes Microsoft Sentinel KQL, Splunk SPL, CrowdStrike Falcon Next-Gen SIEM / LogScale CQL, Sophos cloud and endpoint hunts, shared investigation logic, and two experimental Sigma rules. The PDF and code bundle includes the complete query files and mapping notes.
The Splunk and CrowdStrike examples require a verified mapping from your Entra feeds to the companion's dc_* fields. Those aliases are not native connector fields. Verify both service-principal sign-ins and directory audits, preserve activity and record timestamps, and check export completeness. Endpoint sensors alone do not supply the cloud audit trail. The SPL and CQL templates have offline logic and static checks; they have not been executed in a native engine or customer tenant.
The sections below explain the standalone Microsoft KQL download. Each other query family has its own prerequisites and editor in the companion.
Before you run it
- Use an authorized Log Analytics or Microsoft Sentinel workspace. Confirm access to
AADServicePrincipalSignInLogsandAuditLogs, and verify their ingestion, retention, and tenant scope. - Run one numbered block at a time. The queries do not change the tenant or its objects.
- All times are UTC. The default window begins October 13 and ends at query time. Extend to October 1 where records permit, following ConnectWise guidance. Ensure the portal's time selector does not narrow that window.
- A blank
TenantFilterintentionally searches all tenants visible in the workspace. Supply the Entra tenant GUID to focus one tenant.AADTenantIdis not the Log AnalyticsTenantIdworkspace identifier. - Inspect a known event first. A missing table or query error needs investigation; it is not a zero-hit result. Document the record sources and gaps in the worksheet.
What each block does
01A and 01B: record availability
Count visible records by tenant and day, with first and last timestamps. These are two separate queries so an unavailable source does not silently disappear from a combined result. Empty days are absent from the output. First and last timestamps do not prove uninterrupted collection between them; reconcile with diagnostic settings, ingestion health, exports, and retention.
02: application or address matches
Search the seven vendor-listed Application IDs and three publicly reported addresses. The OR logic keeps activity from a reported address against an unlisted app, and from a listed app at another address, including IPv6. It retains original result, signature, and description fields without a result filter. Verify their actual encoding in your source before classifying success or failure. Authentication success is not confirmed data access or theft. A reported-IP match is a review lead, not a verdict.
Review all available authentication for implicated identities, not only this indicator search. ConnectWise lists all three addresses. Huntress's October 19 update distinguishes application sign-ins associated with 144.172.64[.]128 from the failed user logins it saw at .99 and .60, which it did not believe related. The query preserves the vendor indicator set for investigation; it does not assign the same meaning to every match. See Huntress's update and refresh the indicators against the current advisory. Display names and user agents can assist context but do not establish legitimacy.
03: access-change review
This deliberately broad queue includes application, role, policy, and user-management categories, plus selected operation-name fallbacks. It also includes the generic Audit category documented in the table schema, which can produce a much larger review set. It covers all actors and target identities in those categories, including changes to applications outside the SkyKick list. Verify the category and operation values in your ingestion before narrowing the queue. Authorized administration will also appear.
Each output row represents an audit event and one target, not necessarily a unique change. Use Id and CorrelationId to reconstruct the parent event. The query retains the complete target array and each target's nested modifiedProperties. Do not count expanded rows as incidents.
App and user actor fields are kept separately. An empty field means unavailable in that event, not proof there was no actor. A target object ID is not an application's client ID. Inspect the target object's credential metadata, role assignments, owners, and consent; correlate with approved changes and the audit result. Credential validity dates alone do not establish when the credential was added.
04: recorded credential use
Replace every REPLACE_... value with a verified Entra tenant ID, application client ID, and credential key ID. The placeholder query will return nothing by design; do not interpret that as a negative result. Never paste the secret itself into the query or a shared report.
This pivot looks for authentication records naming the credential. It retains failed attempts too. Check the recorded outcome; a credential match does not by itself establish successful authentication or resource access. If the source records only a certificate thumbprint, adapt the last filter to ServicePrincipalCredentialThumbprint with a confirmed thumbprint. Missing credential identifiers and reuse of previously issued tokens can leave this pivot empty. Use the wider app timeline and available resource records too.
Continue into resource activity
These starters do not query every Microsoft 365 workload. Follow the identity and window into Graph activity logs, if already collected, and available Purview audit records for relevant mail and file activity. Graph logging needs its own configuration and licensing; switching it on does not restore historical requests. Directory audit logs are not a complete read-access ledger. Refer to Microsoft's Graph logging guidance and workload audit activity reference.
Validation and limits
The author and an independent technical reviewer checked the columns and nested structures against Microsoft's published service-principal sign-in and AuditLogs schemas. All five query blocks passed offline syntax and schema-binding checks using Microsoft's Kusto.Language library. Thirty-five independently authored logic and static checks passed, covering app-or-IP matching, IPv6, raw result preservation, time boundaries, tenant scoping, nested targets, missing properties, actor fields, and credential pivots. Invalid syntax and nonexistent-field controls were correctly rejected by the offline checker.
No customer-tenant or Kusto service execution has been performed. Offline parsing and schema binding do not execute the queries; the fixtures exercise intended logic, not a Kusto service or your ingestion. Test these starters on known positive and negative events before operational use. They are an investigation aid, not production-ready detection rules.
Public indicators are included for defensive investigation. Do not contact indicator hosts. Handle exported audit properties and request data as potentially sensitive; redact before external sharing.
Back to the field guide