Scale AI left private emails and confidential AI training materials for Meta, Google, and xAI openly accessible through unsecured Google Docs. Some files were even editable, revealing an alarming lapse in basic data protection.
Business Insider uncovered Scale AI’s mismanagement after reviewing dozens of public Google Docs files tied to high-profile AI projects. Contractors called the system “incredibly janky,” highlighting how the scramble for speed may have come at the cost of security.
Google, Meta, xAI work made public
Scale AI used public Google Docs to manage projects for major clients, including Google, Meta, and xAI, according to Business Insider. The files, some marked “confidential” and editable, were accessible to anyone with the link. Business Insider reviewed thousands of pages across 85 documents containing internal instructions, training data, and project materials.
For Google, at least seven confidential instruction manuals were exposed. These detailed specific issues with Bard — such as difficulty answering complex questions — and guided contractors on how to address them. Anonymization was inconsistent, with some files still showing Google’s branding.
Meta’s files included labeled audio clips for chatbot speech training, and documents outlining expressiveness standards.
xAI’s leaked information described at least 10 generative AI projects, including one named “Project Xylophone,” which featured 700 prompts designed to improve conversation quality across a wide range of topics.
Google and xAI declined to comment, while Meta did not respond to BI’s inquiries.
Personal emails in public view
In addition to data tied to AI companies, Scale AI exposed sensitive information about thousands of its own contractors in public Google Docs. Spreadsheets listed personal Gmail addresses alongside performance labels such as “high quality,” “low quality,” and “cheating.” Files titled “Good and Bad Folks” and “move all cheating taskers” flagged workers for alleged misconduct, while others documented pay disputes and names of contractors who were “mistakenly banned.”
Contractors noted that these documents were easily accessible and, in some cases, editable by anyone with the link.
Scale AI’s response
According to the BI article, a Scale AI spokesperson said, “We are conducting a thorough investigation and have disabled any user’s ability to publicly share documents from Scale-managed systems. We remain committed to robust technical and policy safeguards to protect confidential information and are always working to strengthen our practices.”
Is Scale AI all hype?
Just weeks ago, Meta made its boldest AI bet yet and invested a whopping $14.3 billion in Scale AI, with Mark Zuckerberg handpicking CEO Alexandr Wang to help lead its superintelligence ambitions. But in light of mounting leaks and internal chaos, was the massive investment a massive mistake?
Reuters reports that Google and Microsoft are backing away from Scale AI, though neither has confirmed. OpenAI has been phasing out its involvement with Scale AI for months.
Scale insists nothing has changed on its end; however, with high-profile clients stepping back and reputational damage setting in, the industry may not be as confident.
OpenAI’s not having it. TechRepublic covers Altman’s response to Meta’s eye-popping $100M attempts to lure OpenAI’s top researchers.
