What happened
Replaying the exposed-secrets intrusion reported in November 2026
- 1
Secrets found in public code
From August 2026Threat actors use the ARTEX agentic framework to send large volumes of GitHub queries that pair organization names with secret and cloud keywords, and find application credentials in public code.
- 2
Signing in as the workloads
October 28 to November 3, 2026Threat actors use leaked client secrets to sign in as service principals, from one operator VPS on a hosting provider's network per Todyl, use a DevOps workload identity from the operator's address, and use MSP multi-tenant apps to reach their client tenants.
- 3
Azure secrets and virtual machines
From October 28, 2026Resource Manager operations that return storage keys and other credentials are targeted, Key Vault secrets and storage account data are taken in some cases, and a compromised app invokes Run Command on Azure virtual machines.
- 4
Collection in Microsoft 365
November 2 to 3, 2026The apps search Microsoft Graph across tenants, download SharePoint files that favor credential and configuration names, and access mailbox items.
Service principals with leaked secrets signed in to Microsoft Entra ID and enumerated Azure resources. Microsoft reports Key Vault secrets and storage account data taken in some cases; operations that return credentials were targeted, and Run Command was used on virtual machines with no result stated. MSP apps reached client tenants' files and mailboxes, in at least one case through delegated administration. Refresh token replays and a GDAP automation app were refused. A possibly related password spray also failed; Todyl links it to the operator with low to moderate confidence. No victim is named.
Where cloud-provider controls fit
A Conditional Access policy for workload identities, with named locations that hold your workloads' egress ranges, would have refused the token requests that single-tenant service principals of your tenant made with leaked secrets from the operator's addresses. An application management policy that blocks password addition refuses new client secrets, but a secret that exists keeps working until you remove it. In your subscriptions, a Key Vault firewall that denies other networks would have refused a read of a secret from outside your networks; the sources do not say where the vault reads came from, so that is a condition. Storage key access off, App Service basic authentication off and an Azure Policy deny on Run Command resources would have bounded what the stolen credentials could reach; the sources do not say which credentials were returned. SharePoint application restrictions and Exchange Online application RBAC would have bounded the files and mailboxes the apps reached. A vendor's multitenant app in your tenant stays outside workload identity Conditional Access, and code in a public repository stays with source control, where GitHub push protection blocks a supported secret at the push.
Could this have been prevented?
In hindsight, not as a whole. 2 controls would have refused a step outright, and 10 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.
- 2
- Prevented
- 10
- 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 exposed-secrets intrusion reported in November 2026
1 · Secrets found in public code
1 Contained1 Not covered2 · Signing in as the workloads
1 Prevented4 Contained1 Not covered6 on other clouds3 · Azure secrets and virtual machines
1 Prevented3 Contained6 on other clouds4 · Collection in Microsoft 365
2 ContainedKey facts
- From August 2026 Microsoft observed threat actors using the ARTEX agentic framework to send large volumes of GitHub queries for exposed secrets in public repositories.
- Service principal sign-ins with leaked or stolen application credentials succeeded on dozens of applications across multiple tenants from October 28, 2026, per Microsoft on November 5, and agents searched Azure DevOps for secrets after authenticating; Todyl counted about 412 sign-in attempts from the operator's address between November 1 and 3, 2026.
- As of November 3, 2026 Todyl counted about 45 app registrations targeted at about 18 organizations, more than 33 authenticated, and 27 organizations reached, downstream clients included; it calls the counts a floor, and its update of November 6 gives no newer figure. One multi-tenant app swept 15 client tenants in about 10-second loops that day, and in at least one case access to an MSP was followed within seconds by access to its customers' tenants through delegated administration.
- Microsoft reported Key Vault secrets exfiltrated in some cases, once about five minutes after the GitHub search, and data taken from storage accounts; it says operations that return credentials were targeted, not which returned one. Todyl saw Key Vault tokens requested at its customers and suspects reads it has not confirmed. Run Command was used on virtual machines with no result stated.
- As of November 3, 2026 Todyl counted 29 wildcard Graph searches across about 25 tenants, 81 SharePoint files downloaded and about 2,279 mailbox items accessed. Four AI service tokens and a Partner Center token were issued at MSPs with no use reported; Microsoft says the actor attempted to validate stolen Azure AI resources and abuse is not confirmed.
- Conditional Access blocked every refresh token replay Todyl saw on November 3, 2026, a GDAP automation app was tried against client tenants without success, and every attempt of a possibly related password spray against 16 accounts failed. Todyl links the spray to the operator with low to moderate confidence and saw no new consents, credentials, roles or users as of November 3.
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.Threat actors use agentic AI tooling to find and abuse exposed secretsMicrosoft Threat Intelligence · Nov 5, 2026 · ResearchSecret discovery on GitHub, service principal sign-ins, Azure enumeration, credential-returning operations, Key Vault exfiltration, Run Command, multitenant fan-out, mitigations
- 2.Stolen App Credentials, Real Access: An Actor Is Signing Into Entra ID and Azure DevOps as Your WorkloadsTodyl · Nov 6, 2026 · Vendor reportTimeline, token sequence, MSP fan-out, Azure DevOps federated credential use, token replay, collection counts, Conditional Access blocks, origin assessment
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.