AI Companies Really, Really Want to Buy Your Work Files and Emails

Commercial data brokers are acquiring internal corporate files, employee emails, Slack messages, and call transcripts to train frontier artificial intelligence systems. Brokers offer payouts between $350,000 and $2 million, arguing that public web data is exhausted and private workplace records are essential to advance model capabilities. According to Wired's report, specialized acquisition firms target enterprise communications to capture authentic workflows that synthetic datasets cannot replicate.

This report examines the scale of private data acquisition, the mechanics brokers use to sanitize files, and the legal battles over bankrupt corporate estates. It is written for technology leaders, legal counsels, and knowledge workers whose daily communications are treated as commercial assets. By analyzing broker business models, automated scrubbing limits, and union pushback, this piece explains how workplace communications are commodified across the artificial intelligence sector.

AI Companies Really: what actually changed

The artificial intelligence industry has shifted from scraping public websites to buying private enterprise workflows. Commercial brokers are pitching corporate leaders with offers between $350,000 and $1.8 million, with some reaching $2 million, for documents, emails, customer tickets, and internal Slack communications. Startups like Handshake, Micro1, and Mercor lead this procurement race. Handshake offered a writer $1,000 for story notes and buys graduate coursework, while winding-down specialist Sunset packages, scrubs, and brokers defunct startup archives directly to artificial intelligence laboratories.

New data broker startups are proliferating rapidly across the technology ecosystem to satisfy this surging demand. In a recent Y Combinator batch, 40 of 65 launched startups were classified as data companies. Industry leader Mercor, established in 2023, saw its three cofounders become billionaires at age 22 by supplying proprietary datasets to model builders. While developer Anthropic acknowledged a large data market, it stated it does not buy live workflow from media companies, though brokers maintain authentic workplace history is necessary because simulated workflows lack real business complexity.

AI Companies Really: how it works

Data acquisition begins when brokers sign licensing deals with operating businesses, winding-down startups, or liquidation trustees to extract digital archives. Mercor chief executive Brendan Foody explained that while artificial intelligence companies previously hired contractors to simulate work, those exercises failed to capture the messiness of an operating business. To supply genuine enterprise records, brokers ingest millions of unstructured files, project management tickets, and internal chat threads directly from corporate repositories. These massive archives are processed to teach foundation models how real teams solve operational problems.

Brokers maintain that privacy safeguards are integrated into their pipelines before files reach external buyers. Intermediaries like Sunset and Mercor deploy automated systems to strip out personal identifying information, customer names, and employee identities. Foody stated that this de-identification can be verified and argued that technology has solved the associated privacy challenges. Companies like Tenkara, where commercialization head Jonah Stillman leads data outreach, solicit organizations to acquire operational communications to automate manufacturing back-office logistics. Once scrubbed, these proprietary corpuses are licensed directly to laboratories.

AI Companies Really: why it matters

The financial scale of private data acquisition is expanding to rival physical compute infrastructure across the technology sector. Micro1 chief executive Ali Ansari stated that laboratories, enterprises, and robotics companies currently pay between $40 billion and $50 billion annually to acquire training data. Ansari projected that annual spending on private datasets will exceed $1 trillion within two to four years as competition intensifies. He also predicted that selling proprietary workflow records to artificial intelligence developers will eventually become a more common corporate revenue line item than advertising.

The high value placed on corporate communications was highlighted during bankruptcy proceedings for Spirit Airlines, which put its digital assets up for sale. The auction corpus contained 80,000 corporate email accounts, 500 million internal chat messages, 20.6 million Microsoft SharePoint files, and 1 million employee time-card records. Google initially submitted a $10 million bid for the dataset, but Micro1 escalated the contest by bidding an additional $2.5 million to win the collection. The contest demonstrated that historical enterprise communications have become valuable speculative assets for technology companies.

AI Companies Really: who is affected

Employees and corporate staff are directly affected as organizations monetize internal communications without explicit workforce consent. The proposed sale of Spirit Airlines records was placed on hold after the flight attendants' union objected to the commercial transfer of their private information. The union president noted that employee communications gained commercial value in the artificial intelligence era to train models intended to replace human jobs and commodify labor. Mercor promotes its business under the motto of organizing human intelligence to power the AI economy, raising concerns about replacing the workers who generated the records.

Privacy experts caution that automated anonymization claims cannot fully protect staff members from significant privacy breaches. Mayu Tobin-Miyaji, general counsel at the Electronic Privacy Information Center, filed an amicus brief supporting the flight attendants, explaining that unstructured data in emails and Teams messages is difficult to clean thoroughly. Workers frequently discuss medical appointments, family issues, or sensitive workplace disputes in chats. While Ansari argued that workers will accept data sales due to future abundance, critics counter that employees bear immediate personal exposure while companies capture the financial rewards.

AI Companies Really: what to watch

The coming months will determine whether regulatory agencies and bankruptcy courts place legal limits on the commercial liquidation of employee records. Venture capitalist Gordon Ritter warned that the technology sector has not established appropriate standards for data acquisition and is intentionally moving faster than governance can develop. The resolution of the Spirit Airlines auction will establish an important legal precedent for whether union contracts and worker privacy protections can block bankruptcy trustees from selling communication archives. If the sale proceeds, distressed corporations may routinely liquidate employee correspondence.

Industry watchers should also monitor whether enterprise software customers demand contractual clauses preventing data brokers from harvesting their daily work streams. While intermediaries like Handshake, Mercor, and Micro1 continue expanding, enterprise clients may insist on strict guarantees that prohibit vendors from monetizing proprietary tenant files. Observers must also track whether major developers like OpenAI and Anthropic implement verifiable data provenance standards that reject corpuses gathered without worker consent. As public data sources remain depleted, the balance between aggressive data procurement and workplace privacy will define the operational landscape.