The Account Passed Every Check. The Person Doing the Work Never Took One.

When two reporting fellows at AlgorithmWatch went looking for AI training accounts to buy, they found a market ready to sell them. Sellers advertised on Facebook and recruited through WhatsApp groups, one of them with about 1,000 members, offering accounts on several of the largest contributor platforms. An account registered in the EU was offered for $70 and one registered in India for $25, and one seller claimed to produce ten fake accounts a day, the investigation found. The buyers often hire taskers to do the actual work and split the pay, with the account holder keeping 40 to 50 percent.

The reason is arithmetic. Platforms set pay by a contributor's location and qualification, so an account that passed a qualification test in a high-wage country is worth far more than the labour behind it. "Any western account is of great value as the wages are higher," the reporters wrote.

This is account renting. The platform's records show one qualified annotator with a clean history, and the labels come from somebody the platform has never met. The model trained on that data inherits the gap, because the expert judgment the client paid for came from whoever was working the account that day.

Watching contributors harder has its own cost

Platforms know this, and many have answered by monitoring contributors more closely. That approach is now running into regulators. On October 27 the privacy group noyb filed five complaints in four EU countries against a major data labeling company over how its contributor platform monitors contractors, The Register reported, alleging that the platform tracked device activity outside of work and shared data with third parties. The company called the claims "unfounded and inaccurate" and said its tools are there to verify contributors, protect confidential projects and pay people accurately.

The dispute shows the bind. Watching a contractor's device more closely raises legal exposure in Europe, and it still answers a different question from the one the platform has. Activity monitoring describes a device and a session. A rented account runs on a real device, through a connection the seller arranged to match the account's location, with a skilled tasker producing work that looks right. What the platform needs to know is whether the person working the account today is the person who qualified for it.

Check the person at the moments that matter

We have learned from running HumanCheck for AI data labeling that verification cadence is everything. A check at sign-up confirms who qualified. A check when a contributor starts a high-value task, and again at payout, confirms the same person is still doing the work, because the moment an account changes hands is the moment the sign-up check stops meaning anything.

HumanCheck confirms a real, unique human in under five seconds and gives the platform an anonymous confirmation plus an audit trail it can show its client. Each check answers one narrow question, whether this is the same unique person who earned the account, so a platform can run it as often as the work calls for and keep its contributor experience light.

The results show up quickly. We prevented over $150,000 in fraud loss within the first 3 weeks of going live with one of the biggest AI data labeling companies in the space, and the full case study walks through how they rolled it out.

Where to start this quarter

Put a check in front of payout first, because payout is where a rented account turns into money and where the incentive to hand an account to someone else is strongest. Then add a check when a contributor unlocks higher-paying expert work, since that is the account a seller can charge the most for.

Your clients are paying for the judgment of the people you qualified. Confirming that those people are the ones doing the work is how a platform delivers what it sold. See how HumanCheck works for AI data labeling.