AI data brokers seek workplace files and unpublished reporting notes
Data brokers are approaching businesses and professionals for internal documents, emails and working notes to train AI models. Handshake confirmed an offer for a reporter’s unpublished notes. Companies purchasing the material describe procedures for stripping identifying details, while privacy specialists point to the difficulty of cleaning unstructured messages.
Artificial Intelligence··Evening
Buyers want the work behind finished documents
Data companies are seeking material that shows how professionals reach decisions and carry out their work. Steven Levy’s WIRED investigation describes approaches requesting internal documents, emails and unpublished reporting notes. The proposed payments are offers from intermediaries, rather than verified prices paid in completed sales. Handshake, a company supplying professional data for AI training, confirmed that it had offered a reporter $1,000 for notes. The reporter declined.[1]
Closed businesses become a source of training material
Sunset, a startup buying data from businesses that have closed, says it removes personal information before arranging deals with AI labs. An earlier proposed sale of Spirit Airlines material covered employee time records, email archives and files held in SharePoint. That process was placed on hold after union objections. It is part of the investigation’s background, rather than a newly completed transfer of the airline’s archive.[1]
Privacy claims meet the difficulty of cleaning messages
Brendan Foody, who leads Mercor, describes privacy protections in data handling, as does Micro1’s leader Ali Ansari. Privacy researcher Mayu Tobin-Miyaji points to the difficulty of removing identifying details from unstructured files such as messages. Anthropic told WIRED it does not buy this kind of workflow data from active businesses like WIRED; OpenAI did not respond. These accounts come from a single WIRED investigation.[1]
