Microsoft’s 38TB Oops! How AI Researchers Accidentally Spilled the Beans

In a world where data is the new gold, safeguarding it is crucial. But what happens when the guardians of this treasure make a blunder? Microsoft, a tech giant, recently found itself in hot water when its AI researchers accidentally leaked 38TB of company data. This incident has raised eyebrows and triggered discussions on data security, especially in Artificial Intelligence (AI).

Microsoft's 38Tb Data Leak

The Nitty-Gritty Details

What Exactly Happened?

Microsoft’s AI team was working on training data for image recognition models. They uploaded this data to a GitHub repository and provided a download link via Azure, Microsoft’s cloud storage service. However, the link gave users full access to the Azure storage account. Anyone could view, upload, overwrite, or even delete files.

The Data Involved

The leaked data was not just any data. It included:

  • Full backups of two employees’ computers
  • Passwords to Microsoft services
  • Secret keys
  • Over 30,000 internal Microsoft Teams messages

The Culprit: SAS Tokens

The leak happened due to an Azure feature called Shared Access Signature (SAS) tokens. These tokens are signed URLs that grant access to Azure Storage data. The token could have been set up with limitations but was configured with full access.

The Timeline

Discovery and Action

Cloud security company Wiz discovered the leak on June 22 and immediately alerted Microsoft. Two days later, Microsoft invalidated the SAS token and sealed the leak. An investigation was carried out and completed in August.

How Long Was the Data Exposed?

According to Wiz, the data had been exposed since 2020, making the situation even more concerning.

Implications and Lessons

For Microsoft

Microsoft claimed that no customer data was exposed and no other internal services were at risk. However, this incident serves as a wake-up call for the company to tighten its data security measures.

For the Industry

This case highlights the new risks organizations face when leveraging the power of AI. As engineers work with massive training data, additional security checks and safeguards are essential.


The Microsoft data leak is a cautionary tale for all organizations, especially those diving into AI. It’s a stark reminder that even giants can falter, and the repercussions can be massive when they do.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top