Cloud security firm Wiz disclosed on September 18, 2023 that Microsoft's AI research division had accidentally exposed 38 terabytes of internal data through an overly permissive Azure Shared Access Signature (SAS) token embedded in the public GitHub repository "robust-models-transfer," which hosted open-source models for image recognition. Rather than scoping the token to the intended model files, Microsoft researchers had configured it to grant "full control" permissions over the entire underlying Azure storage account, exposing the personal workstation backups of two Microsoft employees, including passwords to Microsoft services, secret keys, and more than 30,000 internal Microsoft Teams messages from 359 employees. Because the .ckpt model files relied on Python's pickle format, the write access also created a supply-chain risk in which an attacker could have injected arbitrary code into models downloaded by anyone trusting the repository. Wiz reported the issue to MSRC on June 22, 2023; Microsoft invalidated the token on June 24, replaced it on GitHub on July 7, and completed its internal impact investigation on August 16, stating that "no customer data was exposed, and no other internal services were put at risk." The SAS token had first been committed on July 20, 2020 with an expiry that was later extended to October 6, 2051. In response, Microsoft expanded GitHub's secret-scanning service to flag SAS tokens with overly permissive expirations or privileges.

Sources

• -https://www.wiz.io/blog/38-terabytes-of-private-data-accidentally-exposed-by-microsoft-ai-researchers
• -https://techcrunch.com/2023/09/18/microsoft-ai-researchers-accidentally-exposed-terabytes-of-internal-sensitive-data/