A post in a Facebook group promises that you can cash in quickly by teaching Artificial Intelligence models from home:

“Make over $1000 weekely working on remote jobs… Learn how to Get ready to tasks accounts, different profiles, BMS , extensions, free training, proxies… Guarantee earning… 💥💥💯✅” 

This pitch resembles the hook that billion-dollar US companies use to lure data workers into working remote jobs in the online gig economy. 

Similar ads show up in LinkedIn messages or social media feeds, promising flexible hours, no obligations, the chance to join the trail-blazing AI sector, training at no cost, and guaranteed payment.

However, this post includes a spelling mistake (“weekely”) as well as grammar and punctuation errors, which does not usually happen with legitimate recruitment ads.

The image attached to the text shows a screenshot of a dashboard that is associated with an account from Silicon Valley-based Outlier. The company is part of the $13.8 billion data annotation and AI-training company Scale AI. 

Scale AI’s customers have included Europe-based firms such as business consultancy giants Accenture, SAP and Deloitte, US tech giants Meta, OpenAI, Anthropic, and Microsoft, as well as the White House and the US Army. Outlier itself does not reveal clients, but documents indicate that Google and Meta are among them.

Outlier is one of the companies the writer of the post implies they are working for.

By looking through the feed and similar groups, it becomes clear that such posts are not from recruiters. Their authors try to buy and sell accounts for data workers who work for large western data annotation and AI training companies. Not only Outlier accounts are for sale but also ones from similar companies, such as CrowdGen, Echolabs, and Prolific.

AlgorithmWatch kept track of the scammers to find out who was behind such posts, and expose their operations. It turns out that there is an entire black market.

Western wages for the Global South

AI applications are sustained by hundreds of thousands of remote workers who perform manual “training” tasks for a few large companies. For an AI to converse with a human on any topic, it needs training data, such as texts or custom prompts from data workers.

Demand for trainers is so large that companies such as Outlier cast a wide net. They hire workers globally, and pay wages according to what is customary in the respective countries. This leads to a huge difference in pay rates between contractors hired in the Global South and those in Europe and the United States.

To grow their workforce at speed, AI companies rely on automated processes for workers to create accounts on their platforms. Workers submit their name, address and phone number, and verify their phone number via a sign-up link. They provide ID and tax information, and the payment is set up, usually via PayPal. Once an account is verified, it is considered “ready to task” or “ready for work.”

This easy sign-up process allows people in low-income countries to outsmart the system by procuring an account associated with a high-income country. To do this, they need personal data of such a country’s resident.

This is where fraudsters come into play. They buy and sell such accounts on Facebook. On one hand, workers in the Global South, from India or the Philippines, can buy European or US-based accounts and receive a western salary. On the other hand, owners of US- or Europe-based accounts can hire a “tasker” from another country, usually in the Global South, to carry out the work. The account owners and the taskers then split the wage.

Facebook: A marketplace for fraud

AlgorithmWatch looked into the Facebook post cited earlier. It included a WhatsApp link that appears in different profiles across different groups.  

The one who posted it attempts to attract business by showing a screenshot of his earnings. “This is what we call pure happiness,” he says.

Following the offer’s instructions, the link in the post leads to a WhatsApp group called “Remotask Accounts,” named after Remotasks, a sister company of Outlier focused on data annotation, and part of Scale AI. There is no indication that the group is officially connected to Remotasks. 

The messaging group is run by someone with a Kenyan phone number. He claims to have “verified Outlier accounts” that are “ready to task,” based in the USA, Canada, and the Philippines.

We choose a Spanish surname as a pseudonym to interact in the group, “Bartolome,” and use a Spanish phone number. Bartolome is directly contacted by “Felix” who has a Kenyan number and is looking for someone in Spain to set up an Outlier account for him to use (or possibly sell).

Companies like Outlier have an obligation to undertake “KYC” (Know Your Customer) checks on workers performing tasks for them, just as banks check risky customers and try to detect fraud. In general, the companies perform KYC by linking users’ accounts to local phone numbers and addresses as well as account holders’ personal identity and tax number.

Felix needs someone with a Spanish phone number and photo ID to open an account for him. Bartolome has Spanish residency and ID, and Felix asks for his personal details, including the address and phone number. 

Felix may be physically based in Kenya, but to sign up for the Outlier account, he uses a residential proxy in Spain. 

A residential proxy is an alternative to a virtual private network (VPN). VPNs are used to hide real locations online, for example to watch region-locked content on streaming services, by proxying online traffic through IP addresses owned by a VPN service provider. As such addresses are well-known, VPNs are easily identified. Online work platforms are obliged to verify identities and locations. If users of commercial VPNs are detected, they are blocked to prevent fraudulence.

An alternative to VPNs are residential proxies, grey-market services that work like VPNs but make it look like users access a service from a residential address.

Felix hopes to avoid being detected by Outlier’s anti-fraud systems that specifically check for VPN use, and begins opening a new Outlier account with the Spanish residential proxy and Bartolome’s phone number. 

Bartolome waits for a confirmation on his phone. An SMS arrives on the Spanish phone number. Outlier’s verification system asks Bartolome to upload his photo ID. This is where our experiment stops, and we break contact with Felix.

Sham sale of Outlier accounts

The ploy to open fake Outlier accounts by using European phone numbers and IDs appears to be widespread. Felix said that he gets ten false accounts per day ready to task. The group where he found Bartolome has approximately 1,000 members.

Some members of such WhatsApp groups are likely part of a criminal network. Our alter ego Bartolome attempts to purchase an Outlier account. 

“Rehan” who has a Kenyan number offers EU Outlier accounts in Indonesia and the Philippines. He claims to have an account that pays 35 dollars per hour. “Outliers are the best platform now,” he says. Bartolome is interested.

Rehan offers to sell it for 70 dollars with a proxy. He wants to be paid via crypto currency exchange Binance or digital wallet AirTM.

“Just follow my simple steps and you'll earn more under my guidance bro,” he says. On Binance, Bartolome pays 70 dollars into Rehan’s account “Escobar Crypto.” Once the money arrives, Bartolome asks Rehan for the Outlier account and password.

Rehan refuses to give it to him. He says that Bartolome would need a fraud extension, which he could provide for additional 30 dollars. Bartolome transfers the 30 dollars on Binance. But Rehan still holds back the account data. He wants another 50 dollars.

An argument ensues [the chat is edited for clarity]:

Bartolome is not willing to send another 50 dollars to Escobar Crypto, and leaves it at that. But since Bartolome is known to have a Spanish number and ID, another WhatsApp group member asks him to join a credit card fraud operation involving a Spanish hypermarket chain.

Global South workers on the black labor market

Data workers, especially in Europe and North America, receive emails from taskers in the Global South who propose to work on their accounts and give them a cut of the payment in turn. Any western account is of great value as the wages are higher. 

This is a colonial model adjusted in the digital age: European citizens may use their status and residency to be paid for doing nothing, while taskers in the Global South only get a fraction to do all the work. The black market reflects a legal corporate business relationship: US-based companies globally exploit cheap workers’ labor to train their AI models.  

AlgorithmWatch conducted a second experiment. Under the Romanian alias “Tudor,” we published an ad for Outlier taskers in a Facebook group. The company is most commonly mentioned in these groups.

“40pct” means that Tudor is willing to pay 40 percent of the account’s earnings to the tasker.

Members of the Facebook group get in touch with us.

Sanjib from India does not want to be a tasker but offers Tudor access to an Indian Outlier account for 25 dollars. Few people are interested in Indian accounts, as the wages are very low there, but Sanjib is still trying.

Tudor speaks to six others, all from Kenya. The reason so many taskers are from Kenya may be that Outlier’s sister company Remotasks used to operate in the country on a large scale until March 2024, when it shut down all business there. Many people in the east African state have worked on these platforms. All taskers Tudor spoke to say that they can get around the platform’s location blocking by using residential proxies. 

Victor says he trains Large Language Models on English language Outlier accounts. He would apply skills in maths, chemistry, biology, “and more.”  “Am very experienced,” he adds. When Tudor asks for proof, Victor shows an account from Canada he worked from the previous week. “This week no job in it… no tasks added,” he says. “I can do all work in Outlier regardless of where the account is based. Am sure you will like it.”

When Tudor asks him how he managed to circumvent the regional restriction, Victor says: “Its all about quality of the work submitted. Many people are submitting poor quality work leading to suspension of their accounts.” He would not use a VPN. “Proxy is stronger and safer.” Concerning Tudor’s offer, he says: “Not bad although the owner from Canada gives me 50%.”

Jonathan is from Nairobi and his expertise is in mathematical problem-solving. “I mostly do Outlier,” he explains. “I have been tasking on an Italian account before the project was removed.” He provides images for proof, including dates, names of the projects, and the IP address of the Italian residential proxy he uses to access the account.

Tudor also spoke to Mugo from Kenya. He claims to be a “Project lead/Coordinator” at Remotasks, which is unlikely as the company left Kenya. Mugo manages “a team of taskers,” he says, but he is having problems as Outlier keeps flagging their accounts as fraudulent. Here is a transcription of his online dialogue with Tudor (the chat was edited for clarity):

Data workers in Europe have noticed a change in Outlier. The company seems to be getting stricter on suspicious accounts. “They've started banning significantly more people and are actively investigating these scammer rings that people have set up on Facebook,” says an American student in Germany who works for Outlier.

We approached Outlier with questions on whether the company can be certain that tasks performed on the platform are undertaken by the individuals who signed up for the platform, if they are concerned over whether such “taskers” are connected to criminal networks, and whether this could create openings for malicious actors to access sensitive material.

Outlier would not give us direct answers, and instead a spokesperson issued a statement which read: “We take the integrity of our contributor platforms seriously and have strong safeguards in place to prevent fraud. Every contributor must pass identity verification before accessing work, and we employ a layered system of technical safeguards, including real-time behavioral signals and fraud detection. Accounts flagged for suspicious activity are promptly investigated and deactivated as necessary. We are committed to robust systems that reflect industry best practices through continuous testing and improvement to stay ahead of emerging risks.”

Sharing is not caring anymore

When AI companies such as Meta or Google outsource the training of Large Language Models to companies such as Outlier, it leads to an expansion of the subcontracting chain to the black market, called “chain lengthening.” 

The model of sharing accounts has been common in online gig work across the world. Members of a family may work on a single account, with each member completing tasks when they are available. Behind one account can be an entire supply chain of workers. 

Some food couriers and Uber drivers rent their accounts. “It happens especially amongst undocumented migrants, sometimes without a fee, among friends, so those without legal work can get work,” says a gig work specialist who asked to remain anonymous. “It’s commercialised, and sometimes it’s out of solidarity. It seems to be widespread for delivery riders, especially in the UK.” 

In South America, exchanging accounts has been a common practice, says another expert. They can even be considered a form of currency. Platform users may sell a valuable account to pay off debts.

This market has become a blueprint for criminal groups worldwide. Clients do not know anymore who is working on their projects. Our experiment suggests that account managers and taskers may belong to or be attached to criminal networks.

This investigation was conducted in the framework of our Algorithmic Accountability Reporting Fellowship by our reporting fellows Michael Bird and Nathan Schepers.