It turns out that Scale AI, which handles training tasks for Google's AI 'Gemini,' was rampant with poor quality work by unvetted freelancers

AI startup Scale AI has clients including some of the world's leading technology companies, such as Google, Meta, OpenAI, and xAI, and is highly regarded as an 'infrastructure company for the AI era,' growing rapidly through huge funding and strategic partnerships. However, in 2025, Scale AI, which has grown rapidly as a data provider supporting AI development, has been hit by a series of serious problems, and it has been reported that the company's credibility is under serious scrutiny.

Exclusive: Scale AI's Spam, Security Woes Plagued the Company While Serving Google
https://www.inc.com/sam-blum/exclusive-scale-ais-spam-security-woes-while-serving-google/91205895

According to an investigative report by Business Insider , Scale AI uses Google Docs to manage its work, and has published confidential documents related to major clients such as Meta, Google, and xAI online, where anyone with the link can view and edit them. These documents amount to about 85 items, more than 300 pages, and include details of 'Project Bulba,' an internal Google project to improve Bard (now Gemini) using ChatGPT, as well as the names, email addresses, ratings, and work history of freelancers working on contract work.

At Scale AI, work details were recorded in spreadsheet format, with ratings such as 'high accuracy,' 'factual errors,' and 'possible fraud' written along with the names of workers. The problem was that this information was left easily accessible from the outside for a long period of time, and Business Insider pointed out that the scale and duration of the leak were too serious to be called 'unintentional disclosure.' Cybersecurity experts also sounded the alarm, saying that the documents contained information with a high risk of leaking, such as API keys and customer project names, and that this was a very serious management error.

After the leak became public, Scale AI suddenly took steps to set many of its documents to private, but it has already sparked distrust among its biggest clients, with Google planning to terminate its contract with the company, according to Reuters .

The background to these problems is that Scale AI outsourced some of its work to a large number of external freelancers, and there were serious flaws in the management system. According to American media Inc, contractors who had not undergone sufficient screening participated in AI training and evaluation projects for Google, including Gemini, and much of the work submitted was filled with meaningless strings of characters and spam-like content made up of copy and paste. According to reports, some workers were suspected of purchasing other people's accounts to participate in the work, which could have had a negative impact on the learning accuracy of the AI. This proliferation of poor quality data is also seen as one of the factors that prompted Google to decide to terminate the contract.

According to Reuters, Google was set to pay Scale AI more than $150 million in 2024 and was scheduled to sign a contract worth another $200 million in 2025. However, following Meta's investment in Scale AI, Google has decided to terminate the contract. This 'strengthening of the relationship with Meta' represents a conflict of interest with a competitor for Google, so it appears that the company has decided that it would be difficult to continue the deal.

In addition, other tech giants such as Microsoft, OpenAI, and xAI are also reportedly reviewing their relationships with Scale AI, meaning the company faces the possibility of losing a major revenue stream in one fell swoop.

Meanwhile, venture capital firm Accel has invested $350 million in Scale AI so far and owns 18.5% of the company's shares, and is on track to make more than $2.5 billion from Meta's $14.8 billion investment, Bloomberg reports.

Meta in talks to invest 1.4 trillion yen in AI company Scale AI, Meta's largest external AI investment - GIGAZINE

With this investment, Meta has acquired a non-voting controlling stake in Scale AI, effectively becoming a strategic partner.

In connection with this investment, it has been reported that Alexander Wang, founder and CEO of Scale AI, will transfer to Meta and work on the next generation of AI development at the company's newly established AI research lab specializing in 'superintelligence.' Meta has been actively recruiting researchers from other companies such as Google and OpenAI for this lab, and some reports have said that compensation of several billion yen has been offered.

Meta plans to build an AI lab led by Scale AI CEO Alexander Wang specializing in the pursuit of 'superintelligence,' and offers hundreds of millions to tens of billions of yen in compensation to lure AI researchers from OpenAI and Google - GIGAZINE

In addition, on the technical side, Scale AI will release a performance evaluation platform called 'SEAL Leaderboards' in 2024, which will compare and evaluate the performance of LLMs in multiple fields such as mathematics, coding, and language comprehension using private data sets. However, these efforts are also being subject to evaluation in light of the poor information management this time.