Scale AI’s Leaks Put Google and Others at Risk

June 26, 2025

Share this post :

Why in the news?

Scale AI is a data-labelling startup which just leaked data of its contractors through a Google Docs link that can be seen and edited by anyone who has access to the same.

This mishap comes just two weeks after Meta AI invested $14 billion in Scale AI.

Scale AI is a startup that provides companies like Meta, xAI and Google a data centric end-to-end solution to manage machine learning life cycle.

Google’s Data

According to Business Insider’s findings, a few instruction manuals, titled “confidential, were accessible to anyone with the link. Sensitive documents suggested that Google was struggling with its chatbot, then named Bard. It contained information about what Google thought was wrong with Bard and how Scale AI contractors should fix it.

It is suggested that Google used ChatGPT itself to improve its own AI, then named Bard.

Elon Musk’s xAI Data

As of April 2025, Scale AI was running at least 10 generative AI projects for Elon Musk’s xAI.

Among these, “Project Xylophone” was a key initiative: it involved training documents and a list of 700 prompts that showed that the project was designed to help the AI talk better about many different topics, from zombie apocalypses to plumbing.

Meta’s Data

Meta’s training documents, labelled confidential, were also publicly accessible to anyone with the link. They included audio files showing examples of “good” and “bad” speech prompts, highlighting the standards Meta used for expressiveness in its AI systems.

Some projects were aimed to better meta chatbot better to be more conversational and emotionally engaging while dealing with sensitive issues.

Contractor’s Information Accessible to the Public

According to Business Insider, one internal spreadsheet titled “Good and Bad Folks” openly labelled dozens of workers as either “high quality” or suspected of “cheating.”

Another list, disturbingly titled “move all cheating taskers,” contained hundreds of personal email addresses and flagged individuals for “suspicious behaviour.”

Elsewhere, documents revealed detailed breakdowns of how much each contractor was paid, along with internal notes on pay disputes, inconsistencies, and grievances.

The Danger Zone

While handling data of the biggest companies in the world, an organisation should be sensible with the method of securing and safeguarding highly sensitive information of its clients.

Using Google Docs as a method of safeguarding information is unsafe, dangerous.

Data security is one of the major concerns of the century. Organisations, more than ever, need to take crucial steps to safeguard data to stop it from getting into the wrong hands. Clients, too, must remain vigilant: questioning not just what they share and with whom, but demanding to know exactly what shields are in place to guard their data.

In the wrong hands, even the smallest piece of information can lead to serious consequences, but with awareness and action, we can create a safer digital landscape.

Credit for header image: Appslure

[This content is for informational purposes only and does not constitute legal, financial, or investment advice. This has been constituted based on third-party sources. We do not assume any liability for actions taken based on this information.]