According to salespeople who contacted WIRED’s editorial staff, artificial intelligence companies would be willing to pay media companies for their old work documents, emails, and internal messages to train their models. This market, still poorly documented, would go beyond just news data, according to its players: it could impact employees and society as much as the data centers that currently occupy center stage.
- AI companies would pay media firms up to $2 million for their old work documents, emails and internal messages.
- Brokers also buy work files from shutting startups and assignments from master’s students.
- Handshake, Micro1 and Mercor lead this market, which is racing to collect private company data.
- Mercor’s CEO argues that data is as essential to AI as computing power.
The proposals received are priced. A message signed Nick mentions labs like Anthropic and OpenAI paying media companies, including WIRED, “$350K to $1.8M” in a one-time payment for a copy of their old work documents: documents, emails, Slack messages, tickets, etc. Another email, signed Cora, adds call transcripts to the list and raises the possible sum to $2 million. “Public data is exhausted, and labs are spending over $45 billion on private data to build better models,” she writes.
These approaches are part of a broader market that buys the working methods of entire companies. An independent journalist reported receiving an offer of $1,000 from the company Handshake for her work on a single article, which she did not accept. A Handshake representative confirmed to WIRED: “We license professional documents across many domains so labs can test and train how models handle real knowledge work. Journalism is one of those domains.” According to the company, a significant part of its business involves buying the work and assignments of master’s students.
Other companies are positioning themselves in more unexpected niches. An entrepreneur received emails from several companies specializing in helping startups that are shutting down, including one, Sunset, which describes its offer as follows: “We manage the entire process, cleaning and anonymizing the data to remove people from the company and any identifying elements, then negotiating the agreement with the labs.”
Mercor, the flagship
Handshake, Micro1 and Mercor are presented as the leading companies in this sector, which buy up databases “like in a modern gold rush.” Mercor, founded in 2023, is said to be the largest. Last year, its three co-founders became billionaires at age 22, according to Forbes. Like other players, Mercor claims to have partnerships with almost all the major AI labs. Anthropic acknowledges the existence of a vast data market but states it does not buy working methods from active companies like WIRED. OpenAI did not respond before publication.
In an interview, Mercor’s CEO, Brendan Foody, believes the market is set to explode, as the industry is seeking much larger volumes of data. Mercor was originally an AI-focused recruitment company. By interacting with industry players, its team observed that major labs were frustrated with the quality of available data. The industry had previously used experts to simulate real workflows, but this method did not capture the messiness of an active company. According to him, AI has two bottomless needs: “compute and data.” While the industry publicly invests enormous sums in the former, data collection, he asserts, is just as important.
