The secret market for corporate data as AI training fuel

An email recently arrived in the inbox of Steven Levy, a journalist at Wired: "Not sure if you've heard," wrote someone named Nick, "but AI labs like Anthropic and OpenAI are paying media companies such as Wired US$ 350,000 to US$ 1.8 million for a copy of their internal records — documents, emails, Slack messages, support tickets — to train better models." Another email, from someone named Cora, raised the offer to US$ 2 million, adding call transcripts to the list.

This is not an isolated incident. It is a report on an emerging, quiet, and rapidly expanding market: the trade in corporate work records as training data for artificial intelligence. And the most unsettling fact is that the employees who generate those data almost never consented to this use.

What is being sold

What these companies offer is not generic or publicly available data. It is real records of everyday business operations: Slack conversations, emails with managers, Teams meeting transcripts, customer service call records, source code with its comments, project documents with annotations, Jira histories, internal financial reports, CRM profiles. These are the files every organization accumulates over years of operation, and that have now become a commodity.

A data brokerage industry has grown rapidly around this trade. Leading names in the sector, including Micro1, Mercor, and Handshake, make cold calls directly to companies — even newspapers and magazines — offering between US$ 100,000 and US$ 2 million for full access to their operational records. Micro1 announced a US$ 1 billion commitment over 12 months to acquire and license enterprise operational data through its Corporate Data Partnerships program.

The Spirit Airlines case: when a sale becomes a legal battle

The most emblematic case was the attempt to acquire the data of Spirit Airlines, which filed for bankruptcy in May 2026. The digital assets of Spirit included about 100 million emails, 500 million Microsoft Teams items, 20.6 million Microsoft SharePoint files, and 1 million time-card records. Google made an offer of US$ 10 million; Micro1 countered with US$ 12.5 million.

But the sale encountered an unexpected barrier: the flight attendants' union. The union contested the sale, arguing that the files contained employees' personal data — information that, although de-identified, could mention colleagues, managers, and internal contexts in conversations and emails. EPIC (Electronic Privacy Information Center) filed an amicus brief in support of the union. The result was a pause on the deal, creating a precedent that could define the legal boundaries of corporate data commercialization.

Behind the scenes: how the market works

The dynamics behind this trade are simple but unsettling. Data companies like Micro1, Mercor, and Handshake act as intermediaries: they contact companies, negotiate prices, buy the files, remove personally identifiable information, and resell the data to AI labs like OpenAI, Anthropic, and Google. Some brokers also purchase data from defunct companies — like Spirit Airlines itself — that sell all their digital assets during their wind-down phase.

Mercor, founded in 2023, is considered the industry leader: its three co-founders became billionaires at age 22. Brendan Foody, Mercor's CEO, told Wired that AI labs are frustrated with the quality of current data. They hire experts to simulate real workflows, but the simulations fail to capture the complexity and messiness of real work. "AI companies have two bottomless needs — compute and data," Foody said.

Micro1, for its part, estimates that the current corporate data market is worth between US$ 40 billion and US$ 50 billion per year, and that it could exceed US$ 1 trillion in two to four years. Ali Ansari, CEO of Micro1, financed his acquisition program with capital from Citigroup and Hercules Capital. The business is structured to generate recurring revenue for companies that sell their data — which, in theory, incentivizes participation.

The workplace privacy paradox

The central problem is both ethical and legal. When you create an email, a Teams message, or a document during business hours, those data belong to the company, not to you. But what happens when the company decides to sell those records — which contain personal information, opinions, constructive criticism, private conversations — to train AI that could, in the future, replace functions similar to yours?

Labor law experts warn that most companies still do not have clear policies on the fate of employees' data. "Your work email and chat are not a diary," said Jared Pope, employment attorney and CEO of HR platform Work Shield, to Business Insider. "Assume anything you create on a company system could outlive your own job there."

Data de-identification does not eliminate the problem, argues EPIC. Emails and Teams messages often mention names, positions, projects, and contexts that can reveal sensitive information about individuals, even when the formal identification field has been removed.

What comes next

The corporate data market for AI is still in its early stages. But the signals are clear: 40 of the 65 startups in the latest Y Combinator batch are data companies. Offers are increasing. And brokers know the well is running dry — public data from the internet is being exhausted for next-generation model training.

The question is: at what point will the sale of work records to train AI cross the line from legitimate business intelligence to uncompensated exploitation of labor? Spirit Airlines showed that the legal system is still trying to answer that question — and the answer will define the future of work, privacy, and the AI economy itself.

Sources: WIRED, Business Insider, Micro1, Fortune, Forbes

✓ Independent sources cross-checked and verified before publishing