AI Companies Buy Private Work Data for Up to $2 Million as Public Training Sources Run Dry
Updated
Updated · WIRED · Nov 6
AI Companies Buy Private Work Data for Up to $2 Million as Public Training Sources Run Dry
3 articles · Updated · WIRED · Nov 6
Summary
$350,000 to $2 million offers are reaching media outlets and workers as AI labs seek copies of emails, documents, Slack messages, tickets and call transcripts to train models.
Data brokers including Handshake, Mercor, Micro1 and newer startups say public internet data is largely exhausted, pushing labs toward real workplace records that better capture messy human workflows.
$40 billion to $50 billion a year is already being spent on data, Micro1's CEO said, while brokers pitch anonymization as a fix for privacy risks and a future revenue stream for companies.
Spirit Airlines' stalled data sale exposed the legal and ethical fault lines after a union objected to the auction of 80,000 email accounts, 500 million internal messages and other records.
Critics say de-identification may not remove sensitive personal details from unstructured workplace communications, and warn workers' output is being repurposed to build systems that could automate their jobs.