AI Firms Quietly Shopping for Companies' Emails, Files, and Chat Logs

A new kind of sales pitch is landing in inboxes: AI labs are offering media companies and other businesses large one-time payments for their old work records - documents, emails, Slack messages, tickets, and call transcripts - to train better models. One pitch cited fees of $350,000 to $1.8 million, another floated up to $2 million.

The practice is far bigger than it first appears. Data-acquisition firms such as Handshake, Micro1, and Mercor are buying up company workflows at a rapid clip. Handshake, which offered a freelance writer $1,000 for her work on a single story, confirmed it licenses professional documents across many fields, including journalism, and says much of its business is buying graduate students' papers and coursework. Mercor, founded in 2023, claims partnerships with nearly all major AI labs. Anthropic acknowledged a large market for data but said it does not buy workflow from live companies like Wired; OpenAI did not respond before publication.

Privacy is a central worry. Mercor CEO Brendan Foody says personal identifying information is stripped out and the de-identification can be verified. Micro1 CEO Ali Ansari estimates labs, enterprises, and robotics companies spend $40 to $50 billion a year on data and predicts that will exceed a trillion dollars annually within two to four years. Critics are unconvinced. The Electronic Privacy Information Center's Mayu Tobin-Miyaji says emails and chat messages contain unstructured data that is harder to scrub, and the Spirit Airlines flight attendants' union has challenged a proposed sale of the airline's digital assets - including 80,000 email accounts and 500 million internal messages - over private information. That sale is on hold.

The deeper concern, critics say, is that workers are effectively helping train the systems meant to replace them. Ansari predicts selling workflow data will eventually become a bigger revenue line for companies than advertising, and argues the upside is better AI that makes work easier.

Source: Wired