AI Data Brokers Offer Companies Up to $2 Million for Old Emails, Files and Slack Messages

Wired said it had itself received unsolicited pitches. One claimed that labs such as Anthropic and OpenAI were paying media companies between $350,000 and $1.8 million for copies of old work records. Another, which added call transcripts to the list, put the possible payment at $2 million and claimed that "labs are spending over $45B on private data".

The trade matters because data, alongside computing power, is one of the two inputs AI developers say they cannot get enough of. Public text from the web has largely been used up, and the labs want examples of how real businesses operate in order to train models for office work. Its practitioners told Wired the market could affect workers as much as the data centre boom.

The Buyers and Their Pitch

The leading companies are Mercor, Micro1 and Handshake, according to the report. Mercor, founded in 2023, started as an AI recruiting firm; its three co-founders became billionaires at 22 last year. Chief executive Brendan Foody told Wired that the big labs had been frustrated by the quality of data produced by hiring experts to simulate work, which missed the messiness of a real operating business. He said Mercor strips personally identifying information from data that companies stream to it and that the de-identification can be verified.

Micro1 chief executive Ali Ansari told Wired that labs, enterprises and robotics companies pay about $40 billion to $50 billion a year for data, a figure the magazine said it could not verify, and predicted the market would exceed $1 trillion a year within two to four years. Micro1 mostly uses the data it buys to improve its own models but also sells to large AI firms. Ansari said about half of Y Combinator start-ups are data companies; Wired found that 40 of the 65 companies launched in the accelerator's latest batch are classified as data start-ups.

Handshake confirmed it had offered a freelance writer $1,000 for her work on a single story, which she declined. "We license professional documents across a lot of fields so the labs can test and train how models handle real knowledge work," a representative said. Handshake said a large part of its business is buying graduate students' papers and coursework. Newer firms such as Sunset pitch founders on handling the wind-down of failed start-ups, anonymising their data and brokering its sale to labs.

The labs were more guarded. Anthropic told Wired there is a big market for data but said it does not buy workflow data from live companies like Wired. OpenAI did not respond before publication.

The Spirit Airlines Test Case

The fight over the bankrupt Spirit Airlines' digital assets has become the most public test of the practice. Wired put the corpus at 80,000 email accounts, 500 million internal messages, 20.6 million Microsoft SharePoint files and a million time-card records. Google won the auction with a $10 million bid, but Micro1 later offered $2.5 million more, prompting a dispute over the winner, and the sale is on hold after the flight attendants' union objected that it involved members' private information.

More than 100 members of Congress have signed a letter to Google and Spirit asking for payroll records, time cards and some employee communications to be excluded and for limits on how the data can be used, according to KTTC. Google has said it is not buying personal information and that such data will be excluded or de-identified by an independent third party before it receives anything.

Critics doubt that is enough. "It's not clear that anonymizing can fully get rid of privacy issues," said Mayu Tobin-Miyaji, general counsel of the Electronic Privacy Information Center, which has filed an amicus brief supporting the flight attendants. She said unstructured emails and Teams messages can contain employees' discussions of doctor's appointments or family matters that are hard to remove.

The bankruptcy court's handling of the Spirit sale is likely to be the first formal ruling on whether a company's internal communications can be sold for AI training over its employees' objections.