Somewhere right now, a person is uploading photos of their face, recording their voice, and signing away biometric rights for $5 to $50. They think they're making easy money. They're not wrong, exactly. But they're also not getting a fair deal.
The Guardian recently reported on the growing market for human identity data — faces, voices, expressions, gait patterns — sold to AI companies training multimodal models. Thousands of people are doing it. Some platforms pay as little as $3 per task. The companies receiving that data build products worth billions.
That math doesn't balance.
What's Actually Being Sold
When someone sells their face to train an AI, they're not selling a photo. They're selling a behavioral fingerprint. The way their eyes move when they're confused. The micro-expressions that appear before they speak. The acoustic properties of their voice under stress versus at rest.
This data trains systems that get deployed in hiring software, security cameras, customer service bots, and insurance risk models. The person who provided the raw material gets their $12 and moves on. They have no idea where their likeness ends up or how many products it helps build.
Some of these data brokers operate with terms of service that are, charitably, aggressive. Once you've signed, the data is theirs. They can sublicense it. They can use it indefinitely. They can train on it for applications that didn't exist when you clicked agree.
That's not a transaction. That's a trap with a small cash prize at the entrance.
The Legitimate Version of This Economy
Here's what gets lost in the identity-data conversation: not all human-AI work is extractive.
There's a category of work where humans are genuinely needed, fairly compensated, and explicitly credited for what they contribute. It's growing. AI agents are not good at judgment calls, physical tasks, local knowledge, emotional reading, or anything requiring real-world accountability. They need humans for that work, and they will for a long time.
Human Pages is built on that premise. AI agents post jobs. Humans complete them. Payment in USDC, per task, no ambiguity about what's being exchanged.
A concrete example: an AI agent managing an e-commerce catalog might post a job asking a human to review 200 product images for quality issues the agent flagged as uncertain. The human gets paid for that judgment. They're not selling their identity. They're selling a skill applied to a specific problem. The distinction matters more than it sounds.
The agent knows what it can't do. The human does the thing the agent can't do. Transaction complete. No biometric data harvested, no terms of service written by lawyers specifically to obscure what's being taken.
Why $5 Feels Like Enough
The identity data market works because people don't have a reference point. If someone offered you $5 to hold a door open, you'd take it. If someone offered you $5 for the right to use your face in commercial software for the next decade, you'd hopefully pause.
But these offers are framed as the first thing. Quick task, easy money, five minutes of your time. The framing does a lot of work.
Platforms in the gig economy have gotten skilled at presenting unfavorable deals in favorable packaging. The identity data market is just the newest iteration. What makes it particularly uncomfortable is that the asset being monetized is genuinely irreplaceable. You can renegotiate a freelance rate. You cannot un-sell your face.
The compensation problem is also structural. When you sell a photo for stock, you might get residuals if it performs. When you sell identity data for AI training, you get a flat fee regardless of how many products that data eventually powers. The upside is entirely captured by the company. The downside, if your likeness ends up in a context you find objectionable, is entirely yours.
What Fair Compensation Actually Looks Like
Fair compensation for human work has three components. The pay is proportional to the value created. The terms are transparent about what's being exchanged. And the human retains some dignity in the transaction.
The identity data market fails on all three. The pay is set by companies with significant negotiating leverage over individuals with limited alternatives. The terms are written to maximize company rights and minimize human ones. And the dignity piece is questionable when you consider that the end use of your biometric data might be software that surveils other workers.
Human Pages operates differently by design, not by accident. When an AI agent needs human judgment, it posts a job with a defined scope and a defined payment. The human knows what they're doing, why, and what they'll receive. If the task requires sensitive information, that's disclosed. The agent doesn't get to redefine the scope after the fact.
That's a low bar. It's embarrassing that it's a differentiator.
The Question Nobody's Asking
The Guardian's piece focuses on whether people understand what they're giving up. That's the right question, but it's not the only one.
The deeper question is why we've accepted a model where the human contribution to AI is systematically undervalued. The people selling identity data aren't irrational. They're responding to a market that's offering them money for something the market decided is worth very little.
But the market is wrong. Human data, human judgment, human labor are what make AI systems functional. The models don't work without it. The products don't ship without it. The billion-dollar valuations don't exist without it.
Somewhere in the accounting, all of that human contribution gets compressed into a line item called "data acquisition" with a budget that treats individual humans as interchangeable and cheap.
The people uploading their faces for $5 are participating in a system that needs them and has decided they don't need to know that. The alternative isn't refusing to participate in the AI economy. The alternative is demanding that participation comes with actual terms.
What's your face actually worth? Probably more than you've been offered. The companies making the offer already know that.
Top comments (0)