What happened
Replaying the Azure Health Bot June 2024 intrusion / Tenable's research; the run date is not stated
- 1
A redirect reaches the metadata service
Not dated · before Jun 17, 2024A Data Connection follows a 301 redirect from the researchers' host to the metadata service, which answers and issues a token for Azure Resource Manager.
- 2
The token reads an internal subscription
Not dated · before Jun 17, 2024The token lists a subscription internal to Microsoft and hundreds of other customers' resources in it, and the researchers stop there.
- 3
A second redirect in FHIR validation
Not dated · before Jul 9, 2024The FHIR endpoint validator follows a redirect to WireServer and internal AKS components; without control of headers, it gets no direct metadata access.
- 4
Disclosure and fix
Jun 17 to Jul 12, 2024Tenable reports both issues, and Microsoft rejects redirects for data connection endpoints and fixes the FHIR validator.
Through a redirect in Data Connections, Tenable got a valid response from the Azure Instance Metadata Service, obtained a token for Azure Resource Manager and listed an internal Microsoft subscription with hundreds of other customers' resources before stopping; direct requests to internal endpoints were filtered, and lateral movement was not tested. A second redirect, in FHIR endpoint validation, reached WireServer and internal AKS components; it could not set request headers, which limited direct access to the metadata service, and Microsoft stated it had no cross-tenant access. Microsoft fixed both issues, and neither company knows of malicious exploitation.
What the research shows about your Azure workloads
The chain needed only a feature that sends a request to an endpoint its user names. Once a redirect got past the service's own filter, the metadata service and the identity behind the host decided what it reached. If you run a service that fetches URLs for users, reject redirects in its code; the Metadata Security Protocol with a custom allowlist can keep the requesting process away from the metadata service, its token path and WireServer, though it is in preview and its allowlist parameters reached the Azure CLI on May 6, 2025. No Azure setting refuses the reads of a valid Resource Manager token, so the role assignments of the identity behind your metadata service decide what a stolen token lists. Microsoft fixed its service by July 12, 2024, and states no customer action is required.
Could this have been prevented?
In hindsight, no control would have stopped it. 4 controls would have limited how far it went. 4 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.
- 0
- Prevented
- 4
- Contained
- 4
- 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 Health Bot June 2024 intrusion / Tenable's research; the run date is not stated
1 · A redirect reaches the metadata service
2 Contained1 Not covered6 on other clouds2 · The token reads an internal subscription
1 Not covered3 on other clouds3 · A second redirect in FHIR validation
2 Contained1 Not covered4 · Disclosure and fix
1 Not coveredKey facts
- Direct requests from a Data Connection to the metadata service and other internal endpoints were filtered or inaccessible; a 301 redirect from the researchers' host got past the filter and returned a valid metadata response
- Through the same redirect, the researchers obtained a token for management.azure.com and listed the subscriptions it reached, which returned a subscription internal to Microsoft; no source names the identity behind the token
- Listing that subscription's resources returned hundreds of resources belonging to other customers, in Tenable's listing before its June 17, 2024 report; the researchers stopped there and did not test lateral movement
- The FHIR endpoint validator also followed redirects and reached WireServer and internal AKS components; it could not set request headers, which limited direct access to the metadata service, and Microsoft stated it had no cross-tenant access
- Microsoft rejected redirect status codes for data connection endpoints in all regions by July 2, 2024, and fixes for the second issue were available by July 12, 2024; what the second fix changed is not stated
- Microsoft released CVE-2024-38109 for the first issue on August 13, 2024, with a CVSS 3.1 base score of 9.1 (Critical), marked not exploited and fully mitigated; MSRC rated the second issue Moderate on August 5, 2024 and raised it to Important on August 8, 2024
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.Compromising Microsoft's AI Healthcare Chatbot ServiceTenable Research · Aug 13, 2024 · Researchthe Data Connections feature, the metadata filtering, the redirect bypass, the management token, the internal subscription, the cross-tenant resource listing, the halted investigation, the FHIR validation issue, the MSRC timeline, the redirect rejection fix, no known exploitation
- 2.Microsoft Azure Health Bot Server-Side Request Forgery (Data Connection Endpoints)Tenable Research · Aug 13, 2024 · Advisorythe CVE id, MSRC severity, affected versions, the internal subscription that governs customer Health Bot resources, the disclosure timeline, the pause MSRC asked for, the bounty, Microsoft's no-action statement
- 3.Microsoft Azure Health Bot Server-Side Request Forgery (FHIR Endpoint Validation)Tenable Research · Aug 13, 2024 · Advisorythe FHIR validator, the internals reached (WireServer, internal AKS components), MSRC severity changes, affected versions, the disclosure timeline
- 4.Azure Health Bot Elevation of Privilege Vulnerability (CVE-2024-38109)Microsoft Security Response Center · Aug 13, 2024 · AdvisoryMicrosoft's description, CVSS score, exploitation status, full mitigation, no customer action, the release date
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.