What happened
Replaying the Azure service principal takeover / one morning, day not stated
- 1
A leaked app secret signs in
03:27 UTC, day not statedA client secret exposed for about two weeks signs in as the service principal from a hosting provider in the Netherlands.
- 2
Global Administrator in under three minutes
03:28 to 03:50 UTC, day not statedThe application's consented Graph permission makes it Global Administrator, and 26 application registrations get attacker-controlled secrets.
- 3
From the directory to the subscription
04:32 to 04:34 UTC, day not statedElevate access opens the Azure root and an Owner-class role follows. Storage and Event Hub keys are listed, and Key Vault secrets are read.
In one morning the attacker took Global Administrator of the directory, User Access Administrator at the Azure root and an Owner-class role on the subscription, planted secrets on 26 application registrations, listed the keys of four storage accounts and two Event Hub namespaces and read secrets from three Key Vaults. What was read and whether data left the tenant are not stated.
Where Microsoft Entra and Azure controls fit
No Microsoft Entra setting refuses a directory role assignment made by an application you consented to manage roles, and no Azure setting refuses elevate access to a Global Administrator. The grants you consent to decide those two steps. A Conditional Access policy for workload identities would have refused the first sign-in from the hosting provider, if the application is single-tenant and its workload signs in only from known addresses. An application management policy that blocks new client secrets does not refuse a secret that already exists, and the write-up does not say when the leaked secret was made. After the takeover the two policies would only have narrowed the planted secrets, because a Global Administrator can change both. In Azure, an Azure Policy deny on Owner-class role assignments, storage accounts with shared key access off, Event Hubs namespaces with local authentication disabled and Key Vault network rules would have capped what the identity reached, until an Owner changed them.
Could this have been prevented?
In hindsight, not as a whole. 1 control would have refused a step outright, and 7 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.
- 1
- Prevented
- 7
- 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 Azure service principal takeover / one morning, day not stated
1 ยท A leaked app secret signs in
1 Prevented1 Contained3 on other clouds2 ยท Global Administrator in under three minutes
2 Contained1 Not covered6 on other clouds3 ยท From the directory to the subscription
4 Contained1 Not covered3 on other cloudsKey facts
- Sysdig published the analysis on July 14, 2026, with UTC times within one morning but no calendar date and no victim name
- The service principal's client secret had been exposed for about two weeks and replayed from VPN ranges, Tor exit nodes and small hosting providers before the first valid sign-in, from a hosting provider in the Netherlands at 03:27 UTC. Where it leaked was not stated
- The application held the admin-consented Microsoft Graph permission RoleManagement.ReadWrite.Directory, while its roles view showed only Cloud Application Administrator. At 03:29:57 UTC a permanent assignment made outside Privileged Identity Management gave it Global Administrator
- Between 03:29 and 03:50 UTC attacker-controlled client secrets were added to 26 application registrations, and one of them was later reused from a different egress point
- At 04:32:37 UTC the elevate access action gave User Access Administrator at the Azure root scope, and seven seconds later a role assignment gave an Owner-class role on the subscription
- The access keys of four storage accounts and the root keys of the two Event Hub namespaces that carried the audit telemetry were listed, and three Key Vault access policies were changed before secrets were read. Use of the keys, what was read and any exfiltration were not stated
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.No single pane of glass: Anatomy of an Azure permission takeoverSysdig Threat Research Team ยท Jul 14, 2026 ยท ResearchEntry credential, exposure window, directory escalation, persistence on app registrations, elevation to root, subscription role, key harvest, Key Vault access, persistence check
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.