AI firms bid millions for your work emails. What's your price?

AI companies are running out of public data to train their models. Their next target is the data you produce at work.

Google paid $10 million for corporate data from bankrupt Spirit Airlines, according to a report from Business Insider. AI training startup Mercor bid $7.5 million for the same set. The auction is one of the first public signals that companies are willing to pay real money for the internal documents, emails, and chat logs that employees generate daily.

The Spirit data sale is not an isolated experiment. Handshake AI, a training firm, is offering up to $30,000 for what it calls "high-quality written documents" from individuals. The company requires that sellers own the documents and are authorized to share them. The offer raises compliance questions, but it also puts a price on something most workers have never considered selling.

Elon Musk recently told SpaceX employees their data would be used to train Grok, his AI chatbot. "It will inherit your thoughts and ideas," he said at an all-hands meeting. Meta faced pushback last year when it tried to use employee data for training. The pattern is the same: AI companies need workplace-specific data to teach models how professionals actually communicate, negotiate, and make decisions.

Public internet text has been scraped nearly dry. Reinforcement learning from human feedback helped refine chatbot responses, but that method has limits. The next frontier is what the industry calls "workplace grounding" – training models on real workflows, internal jargon, decision trees, and the kind of context that never makes it onto a public website.

This creates a new market. Corporate data that was once considered internal or proprietary now has a buyer. The price varies. A bankrupt airline's records fetched $10 million. An individual's documents might fetch thousands. The range is wide, and no standard valuation exists yet.

Lawyers told Business Insider the trend is a reminder that workplace data is less private than most employees assume. Employment contracts often grant companies broad rights to data created on their systems. When that company goes bankrupt or sells assets, the data can change hands along with the servers.

For workers, the question is whether to participate. Last year, when Meta paid people to smile on camera for AI training, some accepted as little as $25 an hour. Others refused outright. Selling documents is less time-intensive than sitting for hours of recording, but the stakes may feel higher. Handing over internal emails or strategy documents is different from donating a facial expression.

The calculus also depends on what colleagues do. One worker's refusal to sell does not stop a coworker from accepting the same offer. The data set is only as private as its most willing participant.

AI companies are betting that enough people will say yes. The bids for Spirit's data and Handshake's individual offers are early tests of that bet. If the prices hold or rise, more firms will start asking for your work data. The question is whether you have a price.

Alistair Barr's Tech Memo newsletter provided analysis of the reinforcement learning environments used in these training methods.