Scale AI, data-labeling startup at the center of some of the world’s largest artificial intelligence projects, has allegedly left accessible documents and contractor data accessible to the public through unsafe Google Docs link, according to an investigation, according to an investigation, according to a investigation Business Insider. The revelation comes at a time when the company is already under intensive investigation after Mata’s investment of $ 14.3 billion for a 49 percent stake of meta. While Scale AI says it operates independently, customers such as Google and Openai are allegedly reconsidered their relations with the company amidst security concerns.

A Commercial insider The report found that at least 85 Google Docs had publicly accessible to anyone with sensitive materials of thousands of pages who had links. These documents included confidential instructions and internal assessments for alleged AI training projects, including major technical firms such as Google, Meta and Elon’s Musk XAI.

Scale AI, which helps label and process label training for AI models, has come in fire not only to expose the project documents but also to make public the personal data of their contractors. The spreadsheet reviewed by Business Insider showed private email addresses, paid disputes, and even a list of classifying workers by demonstrations, including some accused “cheating”.

Some of the files that have been exposed have details of Google’s efforts to refine their Bard (now Mithun) chatbot using Openai’s chatgpt data, while some others include reaction to Bard’s weaknesses, such as struggling with complex questions. The documents were reportedly “confidential” labeled by Google, but remained open to public access.

In the leaked documents, another confidential XAI project was allegedly sent to Scale AI contractors, called Project Xylophone. It allegedly consisted of 700 signs, which were to help train condensed abilities in landscapes ranging from casual feast to zombie apocalypse. Meta, also, apparently many “confidential” documents were left accessible, including training audio files, which help improve the safety of emotional tones and their chatbot reactions.

A file titled “Good and Bad Folks” openly tagged dozens of contractors as a low quality or flagged them for suspicious behavior.

Another document titled “Move All Cheating Taskrs” has listed hundreds of names and emails, which are flagged for misconduct. Shocking, some of these sheets were edited, which means that any person with URL can add or change entries.

“We are completely investigated and have disabled any user’s ability to share documents from the scal-managed system,” said the AI ​​spokesman said. Commercial insiderThe company said that it is committed to the customer trust and strengthens technical and policy safety measures for data protection.

Some current and former contractors also told business Insider that this system of using shared Google Docs was common on scale and helped freelance workers to speed up operation in their army. Some workers claimed that they still had access to old projects, yet after being re -assigned, documents were continued to receive updates from customers. Some also said that the system is “incredibly janky”.