Skip to main content

Investigating a Cloud Incident with MISP

Azure raises an alert when someone grants a service principal read access to the ledgerarchive storage account. The incident lead needs to know who made the change, what access Azure granted, and whether anyone used it. A Microsoft Entra sign-in helps identify the caller. Two Azure Activity Log records show how the request ended. Blob access requires a different source of evidence.

In this exercise you will import both provider exports into one MISP event, reconstruct the change, and deliver a forensic report with citations back to the records. Case CLD-042, its tenant, accounts, and logs are synthetic. All times are UTC on 20 November 2026.

Collect the two evidence files

Download E1 · Entra sign-in JSON and E2 · Azure Activity Log JSON.

E1 is a saved Microsoft Graph /auditLogs/signIns response with one selected Azure Portal sign-in. E2 is a saved Azure Resource Manager Activity Logs response with two Administrative records under value. The exercise window is 09:00–09:30 in subscription 74097513-65cf-4ee6-84cb-7d485f31e831. The reporting tenant, taken from the collection context, is d78a81b2-00b7-4e59-8e68-178c4a9c3410.

On a real case, export sign-ins from the affected tenant for the relevant time range and collect Activity Log records from the affected subscription, including the opening and closing stages. Save each page and follow any continuation link. Record the collector, tenant, subscription, UTC collection time, time range, filters, and missing pages in the case evidence register. Keep the original bytes in the evidence store. If you select or redact rows for MISP, record the original hash, selection rule, and the new file’s hash. Graph sign-in collection · Azure role change history

Run the following in the directory containing the two downloads. Compare the full digests with the source file objects after import.

sha256sum entra-signins.json azure-activity.json
7a1908b3bf28713d753b684e88151ce770a51f31c21df408fcbf63151e4d7208  entra-signins.json
d6ab8e23d24d7dc8c9ec8a3df9b2c2aaaada55d3cad590503ed9cea843c90fd7  azure-activity.json

Import into one investigation event

The screenshots follow CLD-042 on an isolated MISP 3.0.0 instance in light mode, signed in as a site administrator. MISP 3.0_wip revision 99ebee7 Both provider imports and the cited Markdown download were tested in this interface with PyMISP 2.5.34.4 and misp-modules revision 0803618. The importers read files already collected from the providers. The exercise requires no cloud credentials.

  1. Have a site administrator check that the module service advertises microsoft_audit_import, azure_activity_log_import, and intelligence_briefing_export at GET /modules. Enable import and export services and those modules for the analyst. Update the MISP object templates so file, cloud-account, microsoft-audit-event, azure-activity-log-event, and intelligence-judgment are available. Module configuration · Object definitions used here

  2. As an analyst who can add events, objects, and reports, create CLD-042 | Unexpected blob reader assignment. Set the event date to 20 November, distribution to Your organisation only, and analysis to Ongoing. Leave the event unpublished.

  3. In the event, open Import from (or Import) and select Microsoft Audit Import (microsoft_audit_import). Upload entra-signins.json, enter the reporting tenant above in Optional reporting tenant UUID, and leave Attach the exact source log off. Run the module. In the preview expect one microsoft-audit-event, one cloud-account, and one file. Check the tenant, hash, and sign-in time, then choose Import to save all three objects. Microsoft import formats and settings

    Microsoft Audit Import with the reporting tenant UUID entered, Include_source unchecked, and entra-signins.json selected.

    E1 omits the reporting tenant. Enter its UUID from the collection record and leave the source attachment unchecked.
  4. In the same event, select Azure Activity Log Import (azure_activity_log_import) and upload azure-activity.json. Leave Attach the exact source audit log off. Expect two azure-activity-log-event objects and one more file. Choose Import to save all three, including the opening record. Azure Activity Log import formats and settings

    Import results for event 17 showing the source file azure-activity.json and two azure-activity-log-event objects, with 65 attributes in total.

    The E2 review offers three objects to import. Both Azure records name the same grantee.

Checkpoint. The event should now have six objects and three timed observations. Open the closing Azure observation and follow its extracted-from reference. The comment Azure Activity Log record 2 points to the second member of the file’s value array (value[1]). Check its source object’s SHA-256 against E2. Both importers leave IDS matching and automatic correlation disabled on their attributes. Use the identifiers below to compare observations. Enabling a source attachment would copy the entire uploaded file, including fields omitted from the structured projection, to the event.

If the import reports an error, check the export schema before retrying. Both example files have a value envelope, but their records belong to different APIs. The Microsoft importer also accepts Entra directory audit JSON and Purview CSV with a JSON record in each AuditData cell. Azure Monitor Activity Log exports with a records envelope belong in the Azure Activity Log importer. A next-page warning calls for collecting the remaining pages. Compare native record IDs and source hashes with the saved objects before repeating an upload.

Reconstruct the role change

Open Objects and compare the three observations in order of event-time. On the Timeline → Seen view, their first- and last-seen times mark the same three instants. Azure’s later submissionTimestamp records when each entry became available for querying. Use the provider’s activity time to reconstruct the operation. MISP forensic timelines

UTC Source What the record establishes
09:14:11 E1, record 1 Entra recorded a successful interactive Azure Portal sign-in for alex@tenant.example from 192.0.2.41. The native status.errorCode is zero.
09:16:03 E2, record 1 BeginRequest / Started. The caller requested a role assignment for a service principal. The importer identifies requestbody as the source of the target fields.
09:16:05 E2, record 2 EndRequest / Succeeded, substatus Created. The responseBody supplies the assigned principal, role definition, assignment ID, and storage-account scope.

The Seen tab lists one Microsoft observation at 09:14:11 UTC and Azure observations at 09:16:03 and 09:16:05 UTC on November 20.

The Seen timeline places the sign-in first. The two Azure records follow two seconds apart.

Who made the change? Compare the Entra actor-id with Azure’s caller-object-id. Both are 7ad614f0-90cf-4e61-ad86-1c68e9d46021. Check that Entra’s reporting tenant matches Azure’s caller-tenant-id. The username and IP also agree. Within E2, the two records have distinct native event IDs and share the Azure operation and correlation IDs. Count one role assignment. The Entra sign-in has its own correlation ID. The records identify the same directory account but leave the individual and the precise token used for the write undetermined.

Who received what? In the closing record, authorization-role describes the caller’s User Access Administrator authority. The grantee appears under assigned-principal-id as 6ab25386-9a59-465d-998f-9db68ef01272, with type ServicePrincipal. The role definition ends in 2a2b9908-6ea1-4ae2-8e65-a410df84e7d1, Microsoft’s Storage Blob Data Reader role. Read assignment-scope all the way to storageAccounts/ledgerarchive in rg-ledger. This grant permits reading and listing blobs at that scope. Microsoft role definition

Did anyone read a blob? Neither export covers blob reads. Ask the storage team to check diagnostic coverage from 09:16:05 onward. For example, the following read-only query tests for OAuth-authenticated GetBlob requests from this grantee during the rest of the exercise window, if the account sends logs to Log Analytics.

StorageBlobLogs
| where TimeGenerated between (datetime(2026-11-20T09:16:05Z) .. datetime(2026-11-20T09:30:00Z))
| where AccountName == "ledgerarchive" and OperationName == "GetBlob"
| where RequesterObjectId =~ "6ab25386-9a59-465d-998f-9db68ef01272"
| project TimeGenerated, ObjectKey, StatusCode, ResponseBodySize, RequesterObjectId

Inspect the status and bytes returned for each match. A zero-row result only describes this query, time range, authentication identity, and available logging. In CLD-042 the storage logs and change ticket are outstanding. Record data access and authorization as open questions. StorageBlobLogs fields

Write and export the forensic report

  1. Add an intelligence-judgment object titled Blob read access was granted. State that Azure created the Storage Blob Data Reader assignment for the named service principal on ledgerarchive. Set confidence to high. In confidence-rationale, cite the successful closing record and its agreement with the request. In alternative, record that an approved inventory job could explain the grant and that the change ticket is outstanding.

  2. In Edit relationships, add supported-by from that judgment to each of the two Azure observations. The sign-in provides identity context. The closing response supports the grant itself.

  3. Under Reports, add an Event Report named CLD-042 | Azure role assignment on ledgerarchive with Inherit event distribution. Write for the incident lead. Give the UTC cutoff, tenant and subscription, the three timeline entries, the grant and its scope, and the limits on attribution and data access. Assign the approval check to cloud IAM and the blob-log check to the storage team. In Edit Content, use the object picker to cite the judgment, the three observations, and both source files. Save and open each reference in the preview. Event Report editing and references

    General tab of the CLD-042 Event Report showing the linked judgment, scope, and three timestamped evidence references.

    The saved Event Report links the finding to its judgment and each timeline entry to the observation that supports it.
  4. From the event’s All actions → Download as, choose Intelligence Briefing Export (intelligence_briefing_export). Open the downloaded Markdown and follow J1 to S2 and S3, then compare S3’s UUID with the closing observation in MISP. The supplied export has one judgment, five cited sources, and zero structural issues. Its status is DRAFT because the example has no reviewer approval. For a file marked REVIEWED, follow the fingerprint and Analyst Opinion procedure.

    Download event dialog with the Intelligence Briefing Export action visible at the bottom of the format list.

    Scroll to Intelligence Briefing Export in the event download dialog to get the cited Markdown.
  5. If the incident lead needs a PDF of the narrative, open the saved Event Report → Menu → PDF via print and choose Save as PDF in the browser. GFM simplified is also available for readers outside MISP. Open the exported file as a recipient would and check that the case, timestamps, finding, evidence references, and open questions are legible. Store the approved copy and its file hash in the case record. Review tenant IDs, names, IPs, and scope before releasing a copy to a wider audience. Event Report download options

Read the exported forensic report (Markdown) · Download the completed MISP event (JSON)

To replay the export independently of the MISP interface, run this in the examples directory against your local misp-modules service. The completed event contains the two importer results, the analyst’s judgment and the Event Report. The command saves the file returned by the exporter.

jq '{module:"intelligence_briefing_export", data:.}' investigation.json |
  curl --fail --silent --show-error -H 'Content-Type: application/json' \
    --data-binary @- http://127.0.0.1:6666/query > briefing-response.json
jq '{summary, warnings, error}' briefing-response.json
jq -er '.data // error(.error // "No report returned")' briefing-response.json > report.b64 &&
  base64 --decode report.b64 > CLD-042.md
cmp CLD-042.md forensic-report.md

With the supplied event and module revision, cmp exits successfully without output. A newly authored MISP event has its own identifiers and briefing fingerprint.

Apply the workflow to AWS and Google Cloud

Keep the same evidence register and report structure. Choose the importer by the native record schema.

  • AWS CloudTrail. Use aws_cloudtrail_import for a native Records file, an individual event, or their gzip forms. Extract the native CloudTrailEvent JSON before importing an Event History lookup wrapper. Pivot on eventID, principal or role ARN, recipient account, and resource. An explicit error marks failure. Except for explicit console-login outcomes, the importer leaves an API call without an error at unknown. Verify the state of the resource before reporting a successful change. AWS CloudTrail import guide
  • Google Cloud Audit Logs. Use google_cloud_audit_import with a Cloud Logging LogEntry, a gcloud JSON array, or an entries.list response. For an IAM change, inspect the event’s service, method, and overall status, then follow its linked authorization checks and IAM binding changes. A granted permission check concerns one permission and resource. Use the recorded operation result and binding delta to assess what changed. Google Cloud Audit Logs import guide