As they race to automate white-collar work, AI labs have found the perfect source of training data: your Slack DMs. Business owners are getting six-figure "exploding offers" that expire in 48 hours to sell their data. Instagram is flooded with ads declaring that "AI needs your data" and promising millions. The data is priced by the terabyte, anonymized, and sold to AI giants desperate for an edge.

Tech giants have moved to harvest their employees' data for AI training. And, once focused on dead startups, brokers now pay for live data from active companies. "Right now, the demand is just astronomical," said Dan Novaes, the CEO of Mode. Inc, a startup that sells data to AI model developers. Mundane workplace data is more valuable than ever as OpenAI, Anthropic, Meta, and others build AI agents for businesses.

Annual spending on data for AI is projected to explode from less than $1 billion last year to almost $6.5 billion in 2027, Gartner estimates. The boom has a darker side. Employees rarely profit from — or know about — their Slacks, PowerPoints, and Google Docs being sold to the highest bidder. Suspected hackers have shopped around a large company's stolen codebase for millions of dollars.

Meta's attempt to harvest its own employees' data for its AI models sparked a revolt. This story is part of a series called The AI Data Grab, an inside look at how tech giants are increasingly turning to workers, users, and the physical world to get data to train AI.stories in this series here. Some CEOs are getting blindsided. Doug Kreuzkamp, who runs airline operations company Springshot, was sitting in his office on a Zoom call in August when he found out that Google had bought his data.

Springshot, a longtime vendor for the now-bankrupt Spirit Airlines, helped flights coordinate with workers such as gate agents and cleaners.

"I had less than 48 hours' notice — from the media — that the intellectual property we had spent 15 years building could be given away," Kreuzkamp told Business Insider.

The surging market is pushing into uncharted waters, and the industry risks a reckoning with workers and regulators, said Ifeoma Ajunwa, a law professor at Emory University. The lack of federal privacy laws leaves few legal barriers to companies collecting data from their employees for AI training, she added. "The question is, does the consent of the employees from whom they collected that data actually extend to the sale of the data? I think it's an ethically gray area," Ajunwa said.

The boom has created a thriving group of go-betweens who hunt for small businesses rich in data and low on cash. After scrubbing sensitive information like Social Security numbers and home addresses, they resell it to AI giants. Although the industry is new, it's become highly competitive. Ryan Lock, cofounder of AI broker Polyshares, compares it to the oil boom that hit Texas in the early 1900s, when prospectors snapped up cheap land from farmers who didn't realize the true value of the oil underneath it.

Small-business owners sitting on troves of internal records may not realize the seemingly mundane files could be worth big money. Tactics like 48-hour deadlines have become common, Lock said. "The offers being made look almost identical to those early oil offers," he said from his home base in Dallas. "They'll say, 'I'm offering you double market. I need an answer by Friday.'" Lock and his cofounders went through the data-licensing process themselves with their other companies, sometimes laughing out loud at the lowball offers they got.

Their firm tries to bridge that information gap by giving business owners more transparency about what their data is worth. "There's still plenty of money for everyone, but business owners should receive more of the benefit," he said.

Data that demonstrates how work is actually done is highly valuable, said William Almeida, the chief revenue officer of AI training startup Micro1, because it provides critical context on how decisions are made and products are built. That information can help train agents, the autonomous computer-using AI tools that have become a major focus for companies like Meta and SpaceXAI. As AI companies build new models, the real-world data guides the AI on how best to handle specific tasks.

Micro1 announced a program earlier this year that pays midsize companies $100,000 to $2 million for access to anonymized data. Almeida said AI developers increasingly want "long-horizon" data that can provide a full picture of how work gets done. If a company won't sell its Slack, customer relationship management, or email records, the rest of the corpus is less valuable. The data is priced based on a multitude of factors, with volume being a critical one.

The more white-collar employees a company has, the more valuable its data corpus is, said Novaes, the Mode. Inc CEO. Other factors, like exclusivity, matter — there's lower demand for a codebase that a rival lab already has its hands on. Fields that AI giants are trying the hardest to automate, like law and accounting, see especially high demand. Generally, though, any company — dead or alive, big or small — can sell its data, as long as it has enough.

"People need to wake up to it and take this opportunity," Novaes said. Grant Gooley decided to license his company's data to an AI training company. His agency, MyPortal Marketing, has sold Micro1 everything from Figma and Canva files to checklists for onboarding clients to Google ads. He said it didn't sell client-specific material. "If AI is going to be trained, why not be part of that, and do it responsibly?" Gooley said. While the market creates wealth for some, it's breeding new kinds of cybersecurity and privacy risks for others.

Nick Heiner helps improve AI agents by building office-like digital environments at AI training startup Surge AI. Heiner said that earlier this year, he was approached by a group offering a large company's codebase for millions of dollars. "It became apparent that this group did not have the rights to sell this data," Heiner told Business Insider. "They had illicitly obtained a copy of the codebase and were just going around to all the data vendors saying, 'Can I get a few million bucks for this?'" Security and consent concerns aren't limited to hackers.

Google's bid to buy Spirit Airlines' old data for $10 million has been held up, in part, by flight attendants worried about how their records will be used. Last week, Sen. Elizabeth Warren and over 100 other members of Congress wrote a letter to Google and Spirit urging the companies to protect the privacy of thousands of former Spirit employees.

A Google spokesperson told Business Insider that it's not looking to buy any personal information from Spirit Airlines and that any such data would be either excluded or de-identified by an independent third party. A representative for Spirit said the company is working closely with the flight attendants and other unions to address their concerns.

Former gig workers for Mercor, another AI training startup, have alleged in an ongoing lawsuit that the company encouraged them to upload work materials without their employers' consent. Mercor has denied their claims in court.

The hunger for work records is so strong that some tech giants have tried tracking their own staff. At Meta, employees revolted when the company announced a new program earlier this year to track their keystrokes and mouse movements for AI training. Employees worried that Meta would capture their sensitive data, like credit card numbers and immigration documents.

Although Meta's chief technology officer, Andrew Bosworth, pledged that the data would be "tightly controlled," it turned out that it had been viewable across the company in a major internal leak. The supposedly anonymized data included employees' private conversations and performance records, screenshots obtained by Business Insider show. Meta suspended the project shortly after the breach and hasn't restarted it. Google once considered tracking its staff's AI output and ultimately dropped the idea, Business Insider previously reported.

The Meta revolt hasn't stopped other companies from eyeing employee data as a useful shortcut to AI domination, including the world's richest man. In an all-hands meeting in August, CEO Elon Musk told employees that SpaceX would train its Grok AI on worker data and encouraged them to think of themselves as the model's "parents." "In a way, it will be trained on you. It will inherit your thoughts and ideas and beliefs, and I think that's a good thing," Musk said.

Though the company data market is nascent, the AI training startups cashing in are optimistic. Ali Ansari, Micro1's CEO, says licensing data to labs could become one of the most common revenue sources for businesses worldwide, he recently posted on X. He estimates that, soon, about 10% of the US's GDP could be dedicated to training AI models, thanks to business data and human professionals. Last week, Micro1 announced a commitment to spend $1 billion on company data over the next 12 months.

The growing use of employee records for AI training could heighten the urgency of calls for new data protection and privacy regulations, said Alex Rosenblat, a sociologist and former Uber policy executive. It could also give employees new leverage and negotiating power as the value of their everyday digital footprint grows, she said.

"It's not that data collection wasn't present before; it's that the scale at which it can happen, the scale at which it is useful, has shifted," Rosenblat said. "And that's an opportunity for people to say, 'wait a minute, the value of this is escaping us.'"