A guy in Cape Town filmed his feet while walking to feed seagulls. He made $14. That's half a week of groceries in South Africa, and all he did was record a sidewalk for an app called Kled AI.
Welcome to the gig economy's weirdest new frontier: selling your identity to train artificial intelligence.
The Going Rates
The numbers are real, reported by The Guardian this week:
- Walking videos (filming your daily commute, neighborhood strolls): $14 per task on Kled AI
- Phone microphone access (ambient noise — restaurants, traffic, hotel lobbies): $100/month on Silencio
- Private text messages (your actual chats with friends and family): $0.50/minute on Neon Mobile
- Voice cloning (let anyone use a digital copy of your voice): $0.02/minute on ElevenLabs
- Multilingual conversations: $0.15/minute on Luel AI (Y-Combinator backed)
Jacobus Louw, 27, made $50 in a couple of weeks uploading pictures and videos of his everyday life in Cape Town. Sahil Tigga, a 22-year-old student in Ranchi, India, covers all his food expenses — over $100/month — by letting Silencio record ambient sounds through his phone. Ramelio Hill, 18, in Chicago, sold his private phone chats for a couple hundred bucks. His logic: tech companies already have his data, so he might as well get paid.
Why This Market Exists Now
AI companies are running out of training data. That's not speculation — researchers estimated back in 2022 that high-quality web text would be exhausted by 2026, and here we are. The biggest training datasets (C4, RefinedWeb, Dolma) now restrict AI companies from using them. Labs tried feeding AI-generated synthetic data back into models, but a 2024 Nature study showed that recursive training causes models to collapse into error-filled nonsense.
So the data has to come from real humans. And paying for it is cheaper than getting sued. AI companies know that licensing data from willing participants avoids the copyright lawsuits that scraping the open web invites.
The Catch Nobody Reads the Fine Print For
In 2024, New York actor Adam Coy sold his likeness to Captions (formerly Mirage) for $1,000. His contract said no political use. But once biometric data leaves your phone, enforcement is a different story entirely.
Here's what's at risk:
Deepfakes. Your face and voice, once digitized, can generate synthetic video of you saying anything. The tech to do this is cheap and getting cheaper.
Identity theft. Biometric data isn't like a password — you can't reset your face. If a database gets breached, that's permanent exposure.
Obsolescence. The data you're selling trains models that might replace the very gig work you're doing. You're teaching the machine to not need you.
Scope creep. Most platforms' terms of service grant broad, perpetual licenses. "Training AI" is vague enough to cover almost anything.
Who's Actually Doing This?
Mostly people in the Global South and young gig workers in Western countries. The pay is meaningful in South Africa or India — $14 for a walking video is real money when minimum wage is $1.40/day. In the US, it's beer money. The asymmetry is the whole business model.
Bouke Klein Teeselink, an economics professor at King's College London, told The Guardian this is an emerging category of work that "will grow substantially." He's probably right. The AI industry spent over $100 billion on training in 2025, and human data is the one resource they can't synthesize.
My Take
I get the appeal. If Google already has your search history and Meta has your face from Instagram, why not get $50 for it? But there's a difference between passive data collection and actively licensing your biometrics to companies whose entire business model is making you replaceable.
The 18-year-old selling his texts today might find a synthetic version of himself competing for jobs in five years. That's not sci-fi paranoia — it's the stated goal of the companies buying the data.
Would I do it? For $14? No. But I'm not buying groceries in Cape Town. And that's exactly why this market works.
Top comments (0)