In cities from Cape Town to Chicago, a new and intimate form of piecework is emerging: people are selling the raw materials of their own lives to fuel the artificial intelligence industry. For a few dollars at a time, they are licensing their voices, their faces, their daily walks, and even their private conversations, becoming what experts term the “human infrastructure” of the AI revolution.
The drivers of this global data gold rush are stark. AI companies are facing a critical shortage of the high-quality, human-generated data needed to train sophisticated models like ChatGPT and Gemini. Key datasets are now restricting access, and researchers estimate this publicly available “text mine” could be exhausted as soon as 2026. Relying on synthetic, AI-generated data risks causing models to spiral into error-ridden collapse.
The Human Solution to a Silicon Valley Problem
Enter a burgeoning sector of data marketplaces. Platforms like Kled AI, Silencio, and Neon Mobile act as intermediaries, paying a global network of contributors to fill the data void. For AI companies, this solves two problems: it provides a stream of coveted “gold standard” human data, as noted by AI researcher Veniamin Veselovsky, and it mitigates the legal risks of copyright disputes that come with scraping the open web.
The economics are compelling for the contributors, particularly in regions with high unemployment or devalued currencies. Jacobus Louw, a 27-year-old in Cape Town, earned $50 in a few weeks by uploading videos of his neighbourhood walks to Kled AI. One video alone netted him $14, half a week’s groceries. “As a South African, being paid in USD is more worth it than people think,” he said. For Sahil Tigga, a student in Ranchi, India, recording ambient noise for Silencio brings in over $100 a month, covering his food costs.
Even in wealthier nations, the calculus holds. Ramelio Hill, an 18-year-old welding apprentice in Chicago, made a couple of hundred dollars selling recordings of his private phone chats to Neon Mobile for $0.50 a minute. His reasoning was pragmatic: tech companies already take his data, so he might as well get paid.
A Bargain Struck in the Dark
However, this transaction is shrouded in opacity and laden with long-term risk. Contributors typically grant what Professor Enrico Bonadio, a law expert at City, University of London, describes as “carte blanche” licenses—worldwide, irrevocable, and royalty-free. These allow platforms and their clients to create “derivative works” and use a person’s likeness in perpetuity, with no further payment.
Jennifer King, a data privacy researcher at Stanford University’s Institute for Human-Centered Artificial Intelligence, warns that consumers have little idea how their data will be repurposed. “They’ll have little recourse if so,” she stated. Despite claims of anonymisation, biometric data like voice and face patterns are inherently difficult to strip of identity robustly, raising alarming prospects of use in deepfakes or facial recognition databases.
The risks are not theoretical. In the UK, fraud prevention service Cifas reported over 118,000 cases of identity fraud in just the first half of 2025, with criminals increasingly using AI to forge documents and create fake identities. A recent UK law now criminalises the creation of non-consensual intimate deepfake images, and tech firms face hefty fines for failing to remove such content.
Security failures have already exposed contributors. Neon Mobile’s app was taken offline in September 2025 after a security flaw exposed users’ phone numbers, call recordings, and transcripts. Ramelio Hill, who used the platform, said he was never informed and now worries how his voice might be misused.
Regret and the Illusion of Control
Even with specific contractual safeguards, control can be an illusion. In 2024, New York actor Adam Coy sold his likeness to a company then called Captions (now Mirage) for $1,000. His agreement banned use for political or adult product promotion and was set to expire in a year. Shortly after, friends sent him viral Instagram reels featuring his AI replica posing as a “vagina doctor” promoting unproven supplements.
“It felt embarrassing to explain it to people,” Coy said, adding that the experience has made him wary. He would only consider such work again for “major compensation.”
Kled AI’s founder, Avi Patel, emphasises that his company’s business depends on user trust and that it vets business clients to avoid those with “questionable intent,” such as pornography or government bodies. His firm, backed by investors including Waymo founder Sebastian Thrun, recently raised $5.5 million. Other platforms, like the Munich-based Silencio, which has raised $3.52 million, use blockchain to reward users for noise data.
A Precarious Dead End?
For the workers, the long-term outlook is bleak, according to experts. Professor Mark Graham of Oxford University, author of *Feeding the Machine*, acknowledges the short-term benefit for individuals but warns the work is “structurally precarious, non-progressive and effectively a dead end.” He describes a model reliant on a “race to the bottom in wages” and a temporary demand. Once the data is gathered or demand shifts, “workers are left with no protections, no transferable skills, and no safety net.”
The enduring value, Graham argues, is captured by “the platforms in the global north.” This dynamic is set against a backdrop where AI is already reshaping labour markets. Research from King’s College London’s Professor Bouke Klein Teeselink indicates generative AI is adversely affecting entry-level and some technical roles in the UK.
For now, individuals like Jacobus Louw continue. The income from Kled AI, though erratic, allowed him to save for a $500 course to become a masseur after years struggling with a nervous disorder. He is aware of the privacy trade-off but sees few alternatives. His story, and thousands like it, underscore a stark new reality: as the AI industry’s hunger grows, the most valuable commodity is not silicon, but human experience itself—sold cheaply today, with consequences that may echo indefinitely.
