AI Labs Pay Millions for Private Corporate Data, Sparking Privacy and Labor Disputes

According to WIRED, an emerging data brokerage industry is quietly collecting and selling internal corporate work files, emails, and Slack messages to AI labs like Anthropic and OpenAI for training more powerful models. The report states that these AI labs are willing to pay between $350,000 and $1.8 million as a one-time fee for a copy of a media company's old work records. Another similar pitch raised the potential deal value to $2 million, claiming that "public data is exhausted, and the labs are spending over $45 billion on private data to build better models."

Industry leaders include startups such as Mercor, Micro1, and Handshake. Mercor, founded in 2023, is considered the sector leader, with its three co-founders becoming billionaires last year. Ali Ansari, CEO of Micro1, stated that labs, enterprises, and robotics companies currently pay about $40 to $50 billion annually for data, predicting this will exceed $1 trillion per year within two to four years. Ansari believes that selling workflow data will become a more common revenue line item for companies than advertising.

Although data brokerages claim they strip personally identifiable information (PII) during transmission, privacy advocates and labor unions have expressed concern. During its bankruptcy, Spirit Airlines put its digital assets—including 80,000 email accounts, 500 million internal messages, and 20.6 million SharePoint files—up for sale, triggering a bidding war between Google and Micro1. However, the flight attendants' union charged that the sale involved employees' private information, leading to a hold on the transaction. Mayu Tobin-Miyaji, General Counsel at the Electronic Privacy Information Center (EPIC), noted that unstructured data like emails and Teams messages are difficult to fully anonymize and may leak sensitive content regarding doctor's appointments or family issues.