According to a report, the artificial intelligence startup that recently received a $14 billion investment from Meta has an “incredibly janky” security system in which secret information on clients such as Meta, Google, and xAI is stored in public Google Doc files.
Meta agreed earlier this month to acquire 49% Scale AI shares for $14.8 billion, bringing the startup’s CEO Alexandr Wang onboard to run a new “superintelligence“ lab. That outrageously high price tag suggests that Meta feels Wang and his business are critical to taking the social media giant’s AI section to the next level.
The reports say that the corporation has been unusually loose in its dealings with high-profile clients, leaving top-secret projects and sensitive information such as email addresses and pay details in Google Docs open to anybody with a link. A spokesman for Scale AI shared:
“We are conducting a thorough investigation and have disabled any user’s ability to publicly share documents from the Scale-managed system. We remain committed to robust technical and policy safeguards to protect confidential information and are always working to strengthen our practices.”
What Experts Say Makes Your Business an Easy Target for Hackers?
While there is no evidence that the public files caused a breach, cybersecurity experts believe they could make the organization vulnerable to hackers. Google and xAI did not immediately respond to The Post’s request for comment. Meta declined to comment.
Five current and former Scale AI workers told BI that Google Docs was widely used within the organization. One employee said:
“The whole Google Docs system always seemed incredibly janky.”
It was able to access thousands of pages of project materials from 85 Google Docs that detailed Scale AI’s sensitive work with Big Tech clients, such as how Google fine-tuned its chatbot using OpenAI’s ChatGPT. According to reports, at least seven Google manuals labeled “confidential” and including ideas to improve the chatbot, then known as Bard, were made public.
New Docs Show How Google Teaches Its AIs to Understand Us
According to the reports, public Google Doc files contained information about Elon Musk’s “Project Xylophone,” such as training documents with 700 conversation prompts to develop the conversation abilities of an AI chatbot.
So-called “confidential” Meta training papers, which included audio snippets of “good” and “bad” voice prompts for training its AI products, were also made public. While these covert projects were frequently given codenames, numerous Scale AI contractors claimed it was still simple to determine which customer they were working for.
Some documents associated with codenamed initiatives even inadvertently included the company’s logo, such as a presentation with a Google logo.
When working with AI solutions, the contractors reported that the chatbot would occasionally simply expose the client when asked. The news outlet discovered publicly available Google Docs spreadsheets with thousands of workers’ identities and private email addresses.
One spreadsheet was simply labeled “Good and Bad Folks,” and it classified dozens of employees as either “high quality” or “cheating.” Another document identified workers exhibiting “suspicious behavior.”
More public documents documented individual contractor payments, as well as compensation disputes and abnormalities.