The Azure Health Bot Service, a cloud platform for healthcare organizations to build AI-powered virtual health assistants, was found to have privilege escalation vulnerabilities. Researchers discovered a server-side request forgery (SSRF) vulnerability, CVE-2024-38109, that could allow unauthorized access to cross-tenant resources. By exploiting the platform’s “Data Connections” feature, researchers demonstrated how attackers could bypass mitigations and gain access to Azure’s Internal Metadata Service (IMDS), obtaining tokens to access resources from other customers.
After discovering the SSRF vulnerability, researchers reported the issue to Microsoft’s Security Response Center (MSRC), which began investigating in June 2024. Microsoft swiftly deployed fixes, ensuring that all affected environments were patched by July 2, 2024. These fixes included changes to prevent attackers from exploiting redirect status codes in data connection endpoints, effectively eliminating the attack vector.
In addition to the SSRF vulnerability, a secondary flaw in the validation mechanism for Fast Healthcare Interoperability Resources (FHIR) endpoints was also found during the research. Although the impact of this vulnerability was more limited, it was reported to Microsoft, and fixes were made by July 12, 2024. The researchers respected MSRC’s guidance on responsible disclosure and chose to halt further testing after reporting this issue.
The vulnerabilities identified by the researchers were not tied to the AI models powering the Azure Health Bot Service but rather to the underlying architecture of the service. Despite the seriousness of the issues, Microsoft’s rapid response helped mitigate the risks, protecting healthcare organizations and their data from potential threats.
Reference: