It’s just been revealed that Microsoft researchers accidentally leaked 38TB of confidential information onto the company’s GitHub page, where potentially anyone could see it. Among the data trove was a backup of two former employees’ workstations, which contained keys, passwords, secrets, and more than 30,000 private Teams messages.
- Full control
- Potential disaster
According to cloud security firm Wiz, the leak was published on Microsoft’s artificial intelligence (AI) GitHub repository and was accidentally included in a tranche of open-source training data. That means visitors were encouraged to download it, meaning it could have fallen into the wrong hands again and again.
Stock Depot / Getty Images
Data breaches can come from all kinds of sources, but it will be particularly embarrassing for Microsoft that this one originated with its own AI researchers. The Wiz report states that Microsoft uploaded the data using Shared Access Signature (SAS) tokens, an Azure feature, that lets users share data through Azure Storage accounts.
Microsoft says bizarre travel article was not created by ‘unsupervised AI’
In the age of ChatGPT, Macs are under malware assault
81% think ChatGPT is a security risk, survey finds
Visitors to the repository were told to download the training data from a provided URL. However, the web address granted access to much more than just the planned training data, and allowed users to browse files and folders that were not intended to be publicly accessible.
Sora Shimazaki / Pexels
It gets worse. The access token that allowed all this was misconfigured to provide full control permissions, Wiz reported, rather than more restrictive read-only permissions. In practice, that meant that anyone who visited the URL could delete and overwrite the files they found, not merely view them.
Wiz explains that this could have had dire consequences. As the repository was full of AI training data, the intention was for users to download it and feed it into a script, thereby improving their own AI models.
Yet because it was open to manipulation thanks to its wrongly configured permissions, “an attacker could have injected malicious code into all the AI models in this storage account, and every user who trusts Microsoft’s GitHub repository would’ve been infected by it,” Wiz explains.
The report also noted that the creation of SAS tokens – which grant access to Azure Storage folders such as this one – does not create any kind of paper trail, meaning “there is no way for an administrator to know this token exists and where it circulates.” When a token has full-access permissions like this one did, the results can be potentially disastrous.
Fortunately, Wiz explains that it reported the issue to Microsoft in June 2023. The leaky SAS token was replaced in July, and Microsoft completed its internal investigation in August. The security lapse has only just been reported to the public to allow time to fully fix it.
It’s a reminder that even seemingly innocent actions can potentially lead to data breaches. Luckily the issue has been patched, but it’s unknown whether hackers gained access to any of the sensitive user data before it was removed.