Editorial illustration for AI Firms Quietly Seek to Buy Your Work Emails and Files
AI Firms Buying Your Work Emails and Files
Someone named Nick emailed me recently with an offer. Anthropic and OpenAI, he said, are paying media companies like Wired between $350,000 and $1.8 million for old work records: documents, emails, Slack messages, support tickets. Another pitch, this one from someone named Cora, pushed the number to $2 million and added call transcripts to the shopping list. Her reasoning was blunt: "Public data is exhausted, and the labs are spending over $45B on private data to build better models."
I'd known AI companies were buying specialized data from outside sources, but nobody had covered how big this had gotten. So I started asking around. What I found is an industry that's been moving through company inboxes and internal archives for a while now, buying up the unglamorous paperwork of ordinary jobs, and still hunting for more.
One freelance writer got offered $1,000 by a company called Handshake for her work on a single article. She turned it down. When I called Handshake to check the story out, a representative didn't deny it.
They had plenty to say about why they're doing it.
Probably the most troubling aspect of the data-gathering movement is that, in essence, it enlists workers, not always with explicit consent, to contribute to an effort where the ultimate goal is to replace or automate the human work that goes into all those documents, messages, and emails.
Why this matters
For anyone building or deploying AI systems, this is a reminder that "training data" has quietly become a line item companies negotiate like real estate, not something scraped for free off the open web. If Anthropic and OpenAI are offering $350K to $2 million for Slack messages, call transcripts, and old emails, that tells us the public internet is no longer enough to keep models improving, and proprietary, messy, human workplace communication is the new premium resource. For founders, that's a signal: your company's internal archives might be worth more as a dataset than you think, and the terms of these deals, who owns derivative value, what happens to employee privacy, are being set now, quietly, without much public scrutiny.
For researchers, it raises an obvious question about what "better models" actually means when the training signal is someone else's work emails. We'd like to see these deals disclosed with the same rigor as any other data-licensing arrangement, because right now the market is forming in the dark, and the workers whose Slack messages are being sold have no seat at that table.
Related Reading
Business & Startups Nordic pilot adds Gemini for Education, NotebookLM to boost AI literacy
Business & Startups Kling launches Video O1, all-in-one model with MVL bridge using transformer
Business & Startups DeepSeek Seeks More Capital Weeks After USD 7B Funding Round
From Archives Ramp Data Shows AI Spending Stalling as Anthropic IPO Nears
From Archives OpenAI Could Be IPO Ready in 2027, Says President Greg Brockman
Common Questions Answered
How much are AI companies like Anthropic and OpenAI paying media companies for work emails and documents?
According to recent pitches, Anthropic and OpenAI are offering between $350,000 and $2 million to media companies for access to old work records including documents, emails, Slack messages, support tickets, and call transcripts. These substantial payments indicate that proprietary workplace communication has become a premium resource for training AI models.
Why are AI labs spending over $45 billion on private data instead of using public data?
Public data sources have become exhausted, forcing AI companies to seek private workplace data to continue improving their models. The shift to purchasing proprietary data like emails and Slack messages represents a significant change in how AI firms acquire training material, moving away from freely scraped web content.
What is the ethical concern about AI companies collecting workers' emails and documents?
The most troubling aspect is that workers are often enlisted without explicit consent to contribute data that will ultimately be used to train AI systems designed to replace or automate their own work. This creates a paradox where employees unknowingly help develop technology that could eliminate their jobs.
What does the high price for workplace data reveal about the AI training landscape?
The fact that AI companies are willing to pay hundreds of thousands to millions of dollars for work files and communications demonstrates that training data has become a negotiated commodity rather than freely available information. This shift shows that proprietary, messy human workplace communication is now the most valuable resource for building better AI models.
Further Reading
- AI Companies Really, Really Want to Buy Your Work Files and Emails - WIRED
- Papers with Code - Latest NLP Research - Papers with Code
- Hugging Face Daily Papers - Hugging Face
- ArXiv CS.CL (Computation and Language) - ArXiv
