Before you run
Data & scope
- Required tables
- Time window
- 24 hours; 15-minute groups with at least 25 successful DeleteBlob requests
- StorageDelete resource logs exported to StorageBlobLogs, with at least 24 hours of coverage for investigated accounts.
- Query access to Blob data-plane logs; verify recovery configuration separately before drawing conclusions about permanence.
Documentation review by Codex (AI). No live-tenant execution or detection-performance claim. Validate schema, data coverage and expected results in your workspace.
// Hunting starter. Documentation checked 2026-11-13; not tenant-tested.
StorageBlobLogs
| where TimeGenerated >= ago(1d) and TimeGenerated <= now()
| where OperationName =~ "DeleteBlob"
| where toint(StatusCode) between (200 .. 299)
| extend SafeObjectPath = tostring(split(Uri, "?")[0])
| summarize SuccessfulDeleteRequests = count(),
ObjectEstimate = dcount(iff(isnotempty(SafeObjectPath), pack_array(SafeObjectPath), dynamic(null))),
ObjectSamples = make_set(SafeObjectPath, 5),
RequestSamples = make_set(ClientRequestId, 5),
FirstSeen = min(TimeGenerated), LastSeen = max(TimeGenerated)
by WindowStart = bin(TimeGenerated, 15m), AccountResource = _ResourceId, AccountName,
AuthenticationType, RequesterTenantId, RequesterObjectId, CallerIpAddress
| where SuccessfulDeleteRequests >= 25
| extend IdentityState = iff(isempty(RequesterObjectId),
"OAuth object ID unavailable for this group", "OAuth object ID observed")
| order by SuccessfulDeleteRequests desc, LastSeen desc
| take 200
Read-only query. Run in your own authorized workspace after checking prerequisites.
Use this hunt when a retrieval corpus shrinks unexpectedly or when reviewing a storage cleanup job. It selects successful DeleteBlob requests and groups at least 25 requests within a 15-minute bin. The result is a request-burst inventory, not an assertion of ransomware or irreversible destruction.
Interpret deletion evidence carefully
A successful delete response does not by itself describe the recovery state. Versioning, snapshots, soft deletion, retention and the exact requested object can change what remains recoverable. Confirm those facts through authoritative storage state and approved recovery procedures rather than inferring permanence from an HTTP status.
The query covers the explicit DeleteBlob operation. Container deletion, hierarchical-namespace path operations and other lifecycle mechanisms can require different telemetry and operation names. Validate the actual operation emitted by your workload before using this recipe as a coverage claim. Blob monitoring reference.
ObjectEstimate uses the URI without its query string. Consequently, version- or snapshot-specific requests can collapse onto the same path, and the estimate is approximate. SuccessfulDeleteRequests counts log rows and can include retries or repeated delivery. The two measures intentionally answer different questions.
Resolve the actor without inventing identity
OAuth requester tenant and object fields are retained when available. SAS, account-key and anonymous request representations may not provide those identifiers. Missing OAuth fields must not be interpreted as one shared principal. The grouping also preserves authentication type and reported address, including the schema’s port component. StorageBlobLogs fields.
Compare the burst with the workload owner, retention policy, cleanup schedule and deployment history. A legitimate corpus rebuild can delete many objects, but it still needs a verified reconstruction path. A compromised automation identity may resemble the normal job, so familiarity is not sufficient closure evidence.
Preserve affected-path samples and request identifiers in a restricted case. Determine the full affected set by narrowing the interval and removing sampling limits in an authorized investigation, rather than treating five examples as complete scope.
Tune and validate
Use approved historical cleanup activity to validate operation names, statuses and thresholds. Confirm that failed deletions and unrelated metadata operations are excluded. Do not delete production data to create test signals.
Tune by account, workflow and expected object scope. A low-volume deletion of a critical artifact can fall below this threshold and needs separate monitoring. The hunt scans 24 hours and returns up to 200 groups. It is documentation-checked, has not run in a tenant, and neither executes recovery nor determines whether recovery will succeed.
Verify the source
Microsoft references
Measure normal activity, validate known test cases, account for ingestion delays, choose a threshold and entity mapping, and assign an owner. Results require investigation.
Follow the detection learning path →