○One morning last year, Jacobus Rowe went on his daily walk around his neighborhood to feed the seagulls he spotted along the way. Except this time, he recorded several videos of his feet and scenery as he walked down the sidewalk. The video earned him $14, about 10 times the country’s minimum wage, and the equivalent of half a week’s worth of groceries for 27-year-old Lu, who is based in Cape Town, South Africa.
This video was for an “Urban Navigation” task that Louw found on Kled AI. Kled AI is an app that pays contributors for uploading data such as videos and photos to train artificial intelligence models. Lou made $50 in a few weeks by uploading photos and videos of her daily life.
Sahil Tiga, 22, who lives thousands of miles away in Ranchi, India, regularly makes money by giving Silencio, a company that crowdsources audio data for AI training, access to his phone’s microphone to capture noise from the city around him, such as traffic inside a restaurant or at a busy intersection. He has also uploaded recordings of his voice. Sahil travels to photograph unique environments, such as hotel lobbies not yet marked on Silencio’s map. He earns more than $100 a month from this job. This is enough to cover all food costs.
And in Chicago, 18-year-old welding apprentice Lamelio Hill made hundreds of dollars by selling private phone chats with friends and family to Neon Mobile, a $0.50-per-minute conversational AI training platform. For Hill, the math was simple. Since the tech companies already had so much of his personal data, he figured he might as well get a cut of the profits.
Uploading everything from the scenery around them to photos, videos, and audio of themselves, these gigging AI trainers are at the forefront of a new global data gold rush. As Silicon Valley’s thirst for high-quality, human-grade data exceeds the data that can be collected from the open internet, a booming industry of data marketplaces has emerged to fill the gap. From Cape Town to Chicago, thousands of people are now microlicensing their biometric IDs and intimate data to train the next generation of AI.
But this new gig economy comes with tradeoffs. In exchange for a few dollars, those trainers are fueling an industry that could ultimately make their skills obsolete, while leaving some vulnerable to a future of deepfakes, identity theft, and digital exploitation that they are only beginning to understand.
Keep the AI wheel spinning
Improving AI language models such as ChatGPT and Gemini requires a huge amount of learning material, but they face a lack of data. The most popular training sources, including C4, RefinedWeb, and Dolma, which account for a quarter of the highest quality datasets on the web, are currently restricting generative AI companies from using their own data to train models. Researchers estimate that AI companies will run out of fresh, high-quality text for training as early as 2026. Some labs rely on feeding back synthetic data generated by AI, but such recursive processes can cause models to produce error-filled slops and cause model collapse.
This is where apps like Kled AI and Silencio step in. In these types of data marketplaces, millions of people monetize their identities to feed and train AI. There are many other options for AI trainers besides Kled AI, Silencio, and Neon Mobile. Luel AI, backed by renowned startup incubator Y-Combinator, procures multilingual conversations for around $0.15 per minute. At ElevenLab, we digitally clone your own voice and make it available to anyone for a basic fee of $0.02 per minute.
Buk Klein Teesling, professor of economics at King’s College London, said gig AI training is an emerging job category that will grow significantly.
Tesselinck said AI companies know that by paying to license their data, they avoid the risk of copyright disputes they might face if they relied entirely on content gleaned from the web. AI researcher Veniamin Veselovsky said these companies also need high-quality data to model new and improved behaviors within their systems. “Right now, human data is the gold standard for sampling from outside the model distribution,” Veselovsky added.
Humans who fuel machines, especially those in developing countries, often need money but have few other options for making money. For many gig AI trainers, doing this work is a pragmatic response to economic inequality. In countries with high unemployment rates and depreciating currencies, earning U.S. currency is often more stable and rewarding than local jobs. Some of them struggle to secure entry-level jobs and turn to AI training out of necessity. Even in wealthier countries, the rising cost of living makes selling yourself a logical economic necessity.
However, the pitfalls of gig AI training can be invisible. Some AI marketplaces grant data trainers an irrevocable royalty-free license that allows companies to create “derivative works.” This means that a 20-minute audio recording today can power an AI customer service bot for years to come, and the trainer won’t be paid another cent. Furthermore, the lack of transparency in these markets means that users’ faces can be published in facial recognition databases and predatory advertising on the other side of the world, with virtually no legal recourse.
Louw, an AI trainer in Cape Town, recognizes the privacy trade-offs. And although their income is unstable and not enough to cover all their monthly expenses, they are willing to accept these conditions in order to make money. He suffered from a neurological disorder for years and was unable to find work, but with the money he earned on AI marketplaces like Kled AI, he was able to save $500 to pay for a spa training course to become a masseuse.
“As a South African, being paid in US dollars is more valuable than people realize,” Mr Lowe said.
Mark Graham, professor of internet geography at the University of Oxford and author of Feeding the Machine, acknowledged that money could be meaningful in the short term for individuals in developing countries, but warned that “this research is structurally unstable, unprogressive and effectively a dead end.”
The AI market relies on a “race to the bottom in wages” and “temporary demand for human data,” Graham added. When this demand changes, “workers will have no protection, no transferable skills, and no safety net.”
Mr Graham said the only winner was “the Global North platform”. [that] Capture all the lasting value. ”
full privileges
Hill, a Chicago-based AI trainer, had conflicting feelings about selling his private calls to Neon Mobile. He said he earned $200 in about 11 hours of calls, but the app frequently went offline and sometimes prevented him from releasing past-due payments. “Neon was always shady to me, but I continued to use it to get some easy money to pay bills and other miscellaneous expenses,” Hill said.
Now he’s reconsidering how easy that money was. Neon Mobile was taken offline in September, just weeks after its launch, after TechCrunch discovered a security flaw that allowed anyone to access users’ phone numbers, call recordings, and transcripts. Ms Hill said she had not been informed of this by Neon Mobile and was now concerned about how her voice could be misused on the internet.
Jennifer King, a data privacy researcher at the Stanford Institute for Human-Centered Artificial Intelligence, is concerned that AI marketplaces leave it unclear where and how users’ data will be deployed. He added that without negotiating or knowing their rights, “consumers are at risk of having their data reused in ways they don’t like or didn’t understand or expect, and if that happens, they have little recourse.”
When AI trainers share their data with Neon Mobile and Kled AI, they grant a full license (worldwide, exclusive, irrevocable, transferable, royalty-free) to sell, use, publicly display, store, and even create derivative works from their likeness.
Kled AI founder Avi Patel said the company’s data contract limits use to AI training and research purposes. “The whole business depends on user trust, and if contributors believe their data can be misused, the platform will fail.” He said his company vets companies before selling datasets to avoid working with companies with “questionable intentions,” such as pornography, or “government agencies” that it believes could use the data in ways that violate that trust.
Neon Mobile did not respond to a request for comment.
Enrico Bonadio, a professor of law at City St. George’s, University of London, said these terms allow platforms and their clients to do “just about anything with that material, forever, without additional charge, and without any real way for posters to withdraw their consent or meaningfully renegotiate it.”
More troubling risks include trainer data being used for deepfakes and impersonation. While data marketplaces claim to remove all identifying information such as name and location before selling data, Bonadio added that biometric patterns are inherently difficult to anonymize in any robust sense.
Seller’s regret
Even if AI trainers are able to negotiate more nuanced protections for how their data is used, they may still feel regret. When New York actor Adam Coy sold his likeness to Captions (an AI-powered video editing company now called Mirage) for $1,000 in 2024, his contract guaranteed that his identity would not be used for political purposes or to sell alcohol, tobacco, or pornography, and that the license would expire in a year.
Caption did not respond to a request for comment.
Shortly thereafter, Adam’s friends began forwarding him videos they had found online that featured his face and voice and had millions of views. In one of these videos, an Instagram Reel, Adam’s AI replica claims to be a “vaginal doctor” and promotes unproven medical supplements for pregnant and postpartum women.
“I was embarrassed to explain it to people,” Coy said.
“The comments are weird to read because they’re commenting on my appearance, but that’s not really me,” Coy added. “My feelings [while deciding to sell my likeness] That is, most models were going to scour the internet for data and similar products [anyway]so it might be worth paying the price. ”
Coy said he hasn’t applied for any AI data-related jobs since then. He said he would only consider it if a company offered a significant fee.
