There’s a new black market in the world of AI.
Companies like Scale AI, Surge AI, and Mercor are hiring thousands of contractors around the world to help train AI chatbots by evaluating their responses for Big Tech clients — and that’s given rise to a shadow economy in which workers’ accounts are being illicitly offered for sale.
Business Insider’s investigation found 100 groups on Facebook illicitly selling access to real and fake AI training accounts, which contractors require to work for data-labeling companies. After we flagged the phenomenon, Meta removed about 40 groups and pieces of content for violating its policies and is continuing to investigate, a spokesperson said.
The prominent AI training companies ban account reselling and say they have safeguards in place to prevent the practice. But what are purported to be “verified” accounts continue to be up for sale on Facebook, WhatsApp, and Telegram.
Internal documents show Scale AI, which received a $14 billion investment from Meta in June, has been battling fraudulent and duplicate accounts and VPN misuse for years, and the company has barred users from some countries from projects to curb cheating.
These AI training companies have raised billions this year as tech giants race to secure the data needed to train their AI models and hire contractors, who are often referred to as “ghost workers” for their behind-the-scenes role in AI development.
Business Insider’s findings show how AI training, also known as data labeling, is attracting scammers and shortcut seekers, similar to the account sharing that happened for food delivery and ride-hailing apps. It raises concerns for account buyers and sellers, who could be scammed and have their private information leaked, and for the clients paying the contracting companies.
Sara Saab, vice president of product at Prolific, a UK-based data-labeling startup, said that the company’s research has shown that no single fraud ring is behind the movement, and that the underground industry has reached a level of sophistication seen in bank fraud or concert ticket scalping.
“Technologies that are helping data labeling companies are also helping people with bad intentions, fraudsters and scammers,” she said.
When the work dries up, opportunists step in
Before contractors can receive tasks such as labeling images or suggesting alternative chatbot responses, they have to create accounts and complete screening tests on platforms like Scale AI-operated Outlier or Surge AI-operated DataAnnotation.tech. The work is remote, asynchronous, and pay can hit over $100 an hour.
Platforms like Outlier offer task-based projects that can last anywhere from a few hours to several months, often in regions where language translation or data annotation work is needed, or where pay rates are lower. When projects dry up, so does contractors’ pay. That’s where opportunists step in.
They target would-be trainers by selling “verified” Outlier accounts that belong to people in countries such as the US, where projects are still active, according to two former Outlier contractors in Kenya. Some may be real accounts, and some may be fake — but either way, reselling accounts is barred by the companies.
The two Kenyan contractors, whose identities are known to Business Insider, said they personally know people who had purchased accounts. They added that some also bought accounts if they were unable to pass the screening tests.
Account buyers use tools like a VPN or a “shadow proxy” to mask their true location by routing their internet connection through another person’s device in the target country, said the two former contractors in Kenya.
There are plenty of social media profiles that claim they can teach account buyers how to get around data contracting companies’ rules. Business Insider reviewed YouTube channels and Telegram groups that sell would-be taskers guides to “bypass geo-restrictions” or answers to Outlier’s onboarding tests or specific projects.
Those who “rent” out their accounts to taskers charge an upfront fee, a percentage of future earnings, or both, according to four contractors who have been approached by people looking to buy their accounts.
Business Insider found one paid ad on WhatsApp that was seeking to purchase and resell verified accounts for two of the biggest AI training platforms, Outlier and Mercor.
Playground for scammers
Both sides are wary of scams, two contractors based in the US told Business Insider. Buyers worry that a “seller” may take payment and disappear. Sellers worry that their buyer won’t offer them a percentage of future earnings once they hand over account login details.
In several posts seen by Business Insider, Facebook users said they had been scammed. Some said they paid for an AI training account, only for the scammer to block them and disappear with their money. Others said they were given an email address and password that wasn’t actually registered with an AI training firm.
The two US contractors said that they often received direct messages on Reddit from people seeking to purchase their accounts after they passed the screening tests. Would-be buyers offered them a “fair payment” plus an unspecified cut of future earnings made via the account, according to a chat seen by Business Insider. One contractor said they ignore the requests because their accounts could be banned if they were caught. Another concern: They would be on the hook for income taxes for any paid work performed under their accounts.
Representatives for the biggest data companies, including Scale AI, Mercor, Prolific, and Surge AI, said buying and selling accounts is banned and that they have a variety of mechanisms, from monitoring Facebook groups to account-level pattern analysis, to detect fraud.
Facebook posts advertising the sale of such accounts also violate Meta’s community standards and fraud and scams policy, a spokesperson confirmed.
“We use device, IP, and behavioral safeguards to identify and remove suspicious accounts before they can access any customer work,” said a Scale AI spokesperson.
Scale AI confronts ‘cheaters’
Internal Scale AI documents reviewed by Business Insider show that the company, founded in 2016, has been dealing with scammers for at least two years. On a project for Google in 2024, thousands of taskers were flagged in a document as “suspected spammers” or “cheaters.” In a 2023 spreadsheet titled “Good and Bad Folks” and another called “suspicious non-US taskers,” which Scale accidentally left public to anyone with the link, the company collected details of potential fraudsters.
Another spreadsheet from late 2023 shows rampant use of VPNs and multiple accounts belonging to the same person. The document contains a list of 490 contractors that were removed: 48 for using a VPN and a digital payments app that allows users to withdraw money in US dollars; 70 for accounts registered under the same name; and 11 for having two accounts. Another 21 users were removed for being “low quality” taskers.
The documents also show that a project with an unnamed client was plagued with quality issues. In one project progress tracking document, Scale AI managers discussed various strategies to “be ahead of the spammers.” They included banning certain accounts from Egypt, Kenya, Pakistan, and other countries from participating in the project for using ChatGPT, and blocking the copying and pasting of content. The Kenya-based contractors Business Insider spoke to said that all AI training project opportunities in the region have sharply declined since late 2024.
The black market for Outlier accounts has spurred another cottage industry: hijacking real ones. Contractors have received fake job promotion emails asking for their login credentials. Scale AI banned one user for harvesting workers’ contact info and spamming them in the summer, according to an email seen by Business Insider.
Prolific’s Saab described the targeting of AI training platforms as an “accelerating arms race” between them and fraudsters — and said it requires proactive action from companies to stay ahead.
Have a tip? Contact Shubhangi Goel via email at sgoel@businessinsider.com or Signal at shuby.85. Use a personal email address and a nonwork device; here’s our guide to sharing information securely.
