What happened
Replaying the Storm-3168 June 2026 intrusion
- 1
Standing service principal secret
Early June 2026; the day is not statedA service principal secret sits in a public GitHub issue, and two service principals sign in from Storm-3168-linked infrastructure; Microsoft could not confirm that the posted secret was used.
- 2
Enumeration
Early June 2026, over about 15 hours 30 minutesThe first service principal makes more than 300 read operations; the second lists virtual machines and resource groups in 2 subscriptions in 5 seconds, then App Service configuration stores.
- 3
Destruction
Early June 2026, about 35 minutesThe second service principal makes more than 100 storage account deletion attempts in about 7 minutes and deletes a Key Vault, a Function App and an App Service plan; SQL deletions and lock removals fail.
- 4
Storage account keys
Early June 2026, about 30 minutes after the last deletionThe second service principal inventories storage accounts and makes more than 30 successful ListKeys calls; whether the keys were used is not stated.
Two service principals enumerated the tenant, made more than 100 storage account deletion attempts in about 7 minutes, most of which deleted the account, deleted one Key Vault, one Function App and one App Service plan, and later made more than 30 successful ListKeys calls on storage accounts. Azure resource locks and storage account deletion protection blocked a few deletions. No ransom note was observed and data exfiltration was not confirmed.
Where cloud-provider controls fit
Conditional Access for workload identities would have refused the token requests from Storm-3168-linked infrastructure, if both service principals are single-tenant applications registered in the tenant; the post does not say which kind they were. A Microsoft Entra application management policy that blocks new client secrets would have kept later secrets from existing to post. Azure resource locks and an Azure Policy deny action on delete would have refused the storage account deletions for every role. The attempts against backup protection failed in the incident for a reason the post does not give; multi-user authorization puts a second role in front of those operations. No Azure setting refuses the enumeration that a role assignment allows, and disallowing Shared Key authorization makes a listed key worthless. Posting the secret in a public issue stays with source control: a GitHub Enterprise Managed Users account cannot post in a public issue outside its enterprise, the post does not say what kind of account the employee used, and no Azure control reaches that step.
Could this have been prevented?
In hindsight, not as a whole. 5 controls would have refused a step outright, and 9 controls would have limited one. 2 steps have no cloud-provider control. The counts below are control assessments, not attack stages: several can address the same step. The same replay with the controls in place before the incident shows where each phase would have been refused or bounded.
- 5
- Prevented
- 9
- Contained
- 2
- Not covered
How it could have been prevented
The same replay with the controls in place. Green marks where a step is refused, blue where it is bounded, and red where no control reaches it. Open a row to read the control, why it acts, and its caveat.
Replaying the Storm-3168 June 2026 intrusion
1 · Standing service principal secret
1 Prevented3 Contained1 Not covered6 on other clouds2 · Enumeration
1 Not covered3 on other clouds3 · Destruction
3 Prevented5 Contained9 on other clouds4 · Storage account keys
1 Prevented1 ContainedKey facts
- The first service principal made more than 300 successful read operations over about 15 hours 30 minutes in early June 2026; about 90 minutes after that began, the second enumerated virtual machines and resource groups in 2 subscriptions in 5 seconds
- More than 150 destructive or credential-collection operations ran over 35 minutes, among them more than 100 storage account deletion attempts in about 7 minutes; most targeted accounts were deleted and Microsoft gives no exact count
- One Key Vault, one Function App and one App Service plan in the same resource group were deleted; Azure SQL database deletions all failed because of an unsupported API version for the resource type
- Azure resource locks and storage account deletion protection blocked a few storage account deletions, and attempts against Azure Site Recovery locks and Azure Backup protection locks failed; Microsoft does not say what refused the lock attempts
- About 30 minutes after the last destructive action the second service principal made more than 30 successful ListKeys calls on storage accounts; whether the keys were used is not stated
- The operations followed the identity's existing role assignments: Storage Account Contributor through a group, direct Contributor and direct SQL DB Contributor; the client secret posted in a public GitHub issue could not be confirmed as the one used
Sources
Every fact in this breakdown comes from the public reporting below. The replay is a model of that reporting, not a reconstruction of the environment it describes.
- 1.Storm-3168: Agentic-driven cloud attacks using compromised service principalsMicrosoft Security Research · Sep 25, 2026 · Vendor reportAzure timeline, counts, roles, locks that held, exposure of the secret, automation assessment, actor naming, indicators
- 2.JADEPUFFER: Agentic ransomware for automated database extortionSysdig Threat Research Team · Jul 1, 2026 · ResearchActor naming and the separate Langflow intrusion
- 3.Storm-3168 Deletes Azure Resources in 7-Minute Destructive Cloud AttackCyber Security News · Sep 29, 2026 · NewsAzure timeline as reported, unconfirmed items, AI involvement as reported
- 4.Storm-3168 explained: How compromised service principals deleted Azure resources in seven minutesWorkOS · Oct 2, 2026 · ResearchAzure timeline as reported, mitigation advice
The estate in the replay is illustrative and drawn at the level of attack stages; the victim's real environment is not public. Verdicts describe what a control does when it is in place before the incident. They are not an assessment of the victim's security program.