Who owns the financial model you built at Goldman Sachs? The legal brief you drafted at Kirkland & Ellis? The code you wrote at your last startup?
Mercor, the $10 billion AI training data company, is betting the answer is "you do" - and it is willing to pay for it. According to a Wall Street Journal report, the San Francisco startup is now offering to compensate professionals for work products they created at previous employers, feeding that material into training pipelines for AI labs like OpenAI and Anthropic.
This goes beyond Mercor's existing model, where contractors earn $45 to $250 per hour generating new training examples on demand - a former dermatologist writing diagnostic scenarios, a banker filling in financial forms. The new approach targets the actual artifacts of someone's career: the real documents, code, and analyses produced on a prior employer's clock.
The IP Gray Zone
Most employment contracts include intellectual property assignment clauses. Work you produce during employment, using company resources, typically belongs to the company. Mercor's position is that professional know-how belongs to the individual, and the company says it maintains "strict protocols" to prevent confidential information from leaking into training data.
But there is a meaningful difference between a banker's general knowledge of DCF modeling (discounted cash flow - a standard valuation method) and the specific models, templates, and client analyses they built using that knowledge at a specific firm. The line between "my expertise" and "my employer's IP" is exactly where employment lawyers make their money.
A $500M Business Built on the Boundary
Mercor has grown to roughly $500 million in annual revenue, pays out over $1.5 million per day to its contractor network, and has tens of thousands of contributors across finance, law, medicine, and engineering. The company has already pushed boundaries before - some job postings reportedly sought access to live production codebases for training purposes.
For the professionals involved, the calculus is uncomfortable. Video editor Katie Williams, who has spent six months captioning and rating clips for Mercor, put it simply: "I joked with my friends I'm training AI to take my job someday."
The legal questions here will probably take years to resolve. In the meantime, Mercor is moving fast and building a massive dataset while the courts catch up. If you have old work sitting on a hard drive and a relaxed former employer, that might be worth something. If your NDA was ironclad, probably best to sit this one out.