AI data companies are approaching media outlets and other businesses with offers to buy their internal work records, including documents, emails, Slack messages and support tickets, according to pitches received by Wired. One message named Anthropic and OpenAI as buyers willing to pay media companies such as Wired $350,000 to $1.8 million as a one-time fee for old work records used to train models. Another pitch, which added call transcripts to the list of sought material, raised the figure to $2 million and said public data is exhausted while labs spend over $45 billion on private data.
The companies behind these pitches are part of a broader data acquisition industry that buys the workflows of sizable companies. Handshake offered a freelance writer $1,000 for her work on a single story, and she declined; a Handshake representative confirmed the offer and said the company licenses professional documents across many fields so labs can test and train models on real knowledge work, with journalism among those fields. Handshake said buying graduate students' papers and coursework is a large part of its business.
Other firms are targeting companies that are shutting down. A company called Sunset wrote to an entrepreneur describing a service that helps startups wind down, scrubs and anonymizes data so people and identifying details are removed, then brokers the deal with labs. Industry leaders Handshake, Micro1 and Mercor are racing to acquire databases, with Mercor, founded in 2023, describing itself as the leader; its three cofounders became billionaires last year at 22. Anthropic acknowledged a large market for data but said it does not buy workflow from live companies like Wired, while OpenAI did not respond before publication.
Mercor CEO Brendan Foody said the company began as an AI recruiting firm and learned that big labs were frustrated with data quality, since hiring experts to simulate work did not capture the messiness of an operating business. He said de-identification of personal identifying information can be verified. Micro1 CEO Ali Ansari said labs, enterprises and robotics companies pay about $40 billion to $50 billion a year for data, a figure the article could not verify, and predicted the total would exceed $1 trillion a year within two to four years. Micro1, founded in 2022, uses purchased information largely to improve its own models but also sells to major AI firms.
The contested sale of bankrupt Spirit Airlines' digital assets offered a rare public view of the market. The corpus included 80,000 email accounts, 500 million internal messages, 20.6 million Microsoft Sharepoint files and a million time-card records. Google appeared to win with a $10 million bid, but Micro1 topped it by $2.5 million, setting off a dispute over the winner. The sale is on hold after the flight attendants' union charged that it involved members' private information.
Mayu Tobin-Miyaji, general counsel of the Electronic Privacy Information Center, which filed an amicus brief for the flight attendants, said it is not clear that anonymizing fully removes privacy issues, noting that emails and Teams messages contain unstructured data that is harder to root out, including messages about doctor's appointments or family issues. Venture capitalist Gordon Ritter said norms and standards for data acquisition have not been set and that the AI industry is moving faster than the ability to set them, partly intentionally. The flight attendants' union president said the data became valuable in the AI age to train models to replace jobs and commodify parts of working in an office environment.
Ansari said selling workflow data to AI companies could become more common than ads as a revenue line for companies, and that objections from workers will be manageable because better models will make their lives easier.
More AI news from TechManNews.
