The Hidden Data Gold Rush: How AI Companies Are Buying Your Work History
There’s something strange happening in my inbox these days. Every few weeks, I receive a pitch that feels more like a spy novel than a business proposition. The emails start innocently enough—”Not sure if you’ve heard”—but quickly veer into territory that makes me pause. One writer named Nick recently told me that major AI labs like Anthropic and OpenAI are paying media companies anywhere from $350,000 to $1.8 million for access to their historical work records. Not just published articles, mind you, but the messy raw material of knowledge work: internal documents, email chains, Slack conversations, project tickets, even call transcripts. The pitch is simple: they want to train their models on how real professionals actually think and work, not just what they produce for public consumption.
The implications of this are staggering when you stop to think about it. These companies aren’t just buying data—they’re buying the intellectual DNA of entire industries. They’re purchasing the collective problem-solving patterns, the creative struggles, the collaborative friction that happens behind closed doors. It’s one thing to train an AI on published journalism or public code repositories; it’s quite another to feed it the unfiltered internal monologue of a newsroom deciding which stories to cover, or a design team wrestling with a difficult client, or an engineering group debugging a crisis at 2 AM. That’s the stuff that doesn’t make it into the polished final product, but it’s arguably where the real expertise lives.
When I started digging deeper, I realized this is far bigger than I initially suspected. The people running these data acquisition companies speak about their work with the urgency of pioneers, convinced that they’re sitting on the next transformative force in artificial intelligence. They compare what they’re building to the massive data centers that dominate tech headlines—those enormous concrete structures humming with servers that everyone acknowledges as critical infrastructure. But the data trade, they argue, is just as important and potentially just as consequential for workers and society. The gold rush metaphor gets thrown around a lot, but these people genuinely believe they’re staking claims on something that will reshape how artificial intelligence understands and replicates human expertise.
The Unlikely Middlemen of the AI Era
The companies reaching out to me are part of a booming industry that has quietly positioned itself as the middleman between the people who create valuable professional work and the AI labs desperate to consume it. There’s a certain irony in the timing of these pitches—we’re all watching the public battles over copyright and creative ownership, the lawsuits from authors and artists, the heated debates in newsrooms about whether AI companies should be allowed to train on published work. But underneath all that very public conflict, there’s a parallel economy operating with much less scrutiny. It’s an economy built on the idea that if public data is exhausted, the next frontier is the private, messy, unglamorous data that real professionals generate simply by doing their jobs.
A freelance writer recently shared her experience with a company called Handshake, which offered her a thousand dollars for the work she’d done on a single story. She declined, but the offer itself reveals how this market operates. When I spoke with a representative from Handshake, he confirmed that the company is actively licensing professional documents across numerous fields so that AI labs can test and train their models on real knowledge work. Journalism, he told me, is just one sector among many. What’s particularly interesting is that Handshake also acquires graduate student papers and coursework—the raw output of the next generation of professionals, captured before they’ve even fully entered their fields.
My entrepreneur friends are getting pulled into this orbit too. They’ve been receiving emails from companies offering to help them wind down their startups in ways that sound almost too convenient. These firms promise to handle the entire process—scrubbing and anonymizing all the company data, stripping away identifying details, and then brokering a deal with the AI labs. It’s a strange afterlife for a failed startup’s digital remains, but it’s becoming a viable exit strategy for founders who might otherwise walk away with nothing. The data they accumulated over years of building—the product specs, the customer conversations, the strategic pivots, the internal debates—becomes a salable asset that benefits from the AI industry’s insatiable hunger.
The Newest Players in a Young Industry
The most striking thing about this data acquisition boom is how young everyone involved is. The industry leaders—Handshake, Micro1, and Mercor—were practically born yesterday in business terms, but they’re already dominating the field. Mercor, the apparent frontrunner, was founded in 2023, and by last year its three cofounders had become billionaires at the age of twenty-two. That kind of meteoric rise would be remarkable in any industry, but it’s particularly noteworthy here because it demonstrates just how desperate the AI labs are for what these companies are selling. The fact that college-age founders can build billion-dollar businesses brokering other people’s work data tells you something about the scale of the opportunity.
When I spoke with Brendan Foody, Mercor’s CEO, he explained that the market is poised to explode as the AI industry’s data requirements grow exponentially. Mercor started as an AI recruiting company, but as the founders got to know the major players in the field, they noticed a persistent frustration. The big labs were unhappy with the quality of the data they were getting. The industry standard had been to hire experts to simulate actual professional workflow—to essentially act out what a knowledge worker does so that AI models could learn from those performances. But simulations, no matter how well executed, can’t fully capture the chaotic reality of an operating business. The economists might describe it as a difference between the laboratory and the field, but really it’s about the fundamental difference between performing work and actually doing it.
Foody urged me to think about Big AI as having two bottomless needs: compute power and data. The industry has been very publicly investing gigantic sums in the former—announcing billion-dollar data centers, forging deals with energy companies, building out physical infrastructure that can be seen from space. But the data gathering, he emphasized, is just as important and operates with much less fanfare. It’s a quieter revolution happening in the background, with less glamour but potentially equal significance. Every piece of professional knowledge that gets captured and fed into these models is shaping what AI will be able to do in the future, and right now, that capture is happening through this shadowy network of brokers who see value in the digital debris of ordinary work lives.
The Valuation of Work’s Hidden Byproducts
What’s particularly fascinating about this data economy is how it forces us to reconsider what professional work is actually worth. When a freelance writer produces a story, they’re paid for the finished product—the article that gets published, the reporting that gets credited to their name. But that same work generates enormous amounts of invisible byproducts: the interview notes, the dead ends and false starts, the multiple drafts, the email exchanges with editors, the research that didn’t make it into the final piece. For decades, this behind-the-scenes material was essentially worthless to anyone except the writer themselves, serving as potential material for future stories or as documentation of their process.
Now, suddenly, that detritus has a market value. AI companies have realized that the real expertise of professionals isn’t just in their final outputs—it’s in their process. The way a journalist evaluates sources, the way a coder approaches a bug, the way a designer iterates toward a solution—these patterns are as valuable as the end results, perhaps more so, because they reveal the underlying logic of skilled work. When models are trained on finished products, they learn to mimic the appearance of expertise. When they’re trained on the messy process, they might actually learn to think like an expert. That’s the bet these companies are making, and it’s why they’re willing to pay millions for access to raw workflow data.
The ethics of this trade are complicated, to say the least. On one hand, there’s something democratizing about the idea that the value of professional knowledge should be recognized and compensated, even in its raw form. On the other hand, there’s a troubling dimension to this data being stripped of its human context—anonymized, scrubbed, and fed into models that will eventually be used to automate the very kinds of work that created the data in the first place. The journalist whose reporting process gets absorbed into an AI model might be contributing to the technology that eventually replaces them. The graduate student whose coursework gets licensed might be funding the system that makes their future job search harder. There’s a paradox inherent in this arrangement that nobody seems to have fully grappled with.
The Existential Questions Beneath the Market
What makes this data trade particularly challenging to wrap my head around is how it intersects with broader questions about labor, value, and technological progress. The professionals whose work is being captured—journalists, engineers, designers, academics—are being paid for their output, but they didn’t sign up for their process to become a training resource. They weren’t consulted about having their internal communications, their struggles, their failures, their intellectual detours packaged and sold to the very companies that might eventually make their expertise obsolete. There’s something almost vampiric about the arrangement, as if the AI industry recognized that the most valuable thing it could consume was not the products of human labor but the labor itself—the hidden cognitive work that produces those products.
The companies doing this acquisition frame themselves as brokers of opportunity, connecting the creators of valuable knowledge with the technologies that will amplify it. But the power dynamics are decidedly unequal. Individual workers rarely benefit directly from these transactions; it’s their employers, their schools, their corporate entities that control the data and make the deals. The writer offered a thousand dollars for her work process represents the high end of what individuals might receive, but most people will never know that their work contributed to an AI model’s training data. They’ll just notice, eventually, that the tools they use seem to understand their jobs disturbingly well.
There’s also a question about what this means for the future of expertise itself. If AI models are trained on the accumulated work processes of millions of professionals, they’ll eventually be able to replicate not just the outputs but the thinking of experts across countless fields. That could mean something as mundane as better autocomplete for professional writing, or something as profound as AI systems that can genuinely reason about complex problems with the depth of human experts. The optimists see this as democratization—making expert knowledge available to everyone. The pessimists see it as the commodification and eventual replacement of human expertise. The truth, as always, probably lies somewhere in between, but the pace of change is so rapid that we’re making these consequential decisions about our collective intellectual future without much public debate.
The Quiet Consolidation of Intellectual Capital
What strikes me most about this emerging economy is how quietly it’s operating. There’s no congressional testimony, no dramatic public hearings, no press conferences about the ethics of buying human work processes for AI training. Instead, it’s happening through cold emails and anonymous brokers, through startup wind-down services and graduate student paper purchases. The industry is being built in the gaps between legal frameworks that were designed for a different era, exploiting ambiguities about who owns the byproducts of intellectual work and what constitutes fair use of professional knowledge.
The data acquisition companies are careful to present themselves as legitimate businesses meeting a genuine market need. They offer anonymization, they emphasize their partnerships with major labs, they position themselves as essential infrastructure for the AI revolution. And in some ways, they’re right. The AI industry does need data, and this is a way to get it that respects at least some property rights. But the deeper question is whether this is how we want our collective intellectual inheritance to be handled. Every era has its resources that define its economy—coal and steel, then oil and silicon, now data. The difference is that this resource is not extracted from the earth but from human minds, and the people doing the extracting aren’t the ones who did the thinking.
As I receive more of these emails and watch this industry grow, I’m struck by how unprepared we are as a society for the questions it raises. We haven’t decided who truly owns professional knowledge, who should benefit from its transformation into training data, or what obligations the collectors have to the people whose work they’re capturing. The companies are moving fast, confident that their innovations will outpace any regulatory response. They might be right. By the time we’ve figured out how to think about this, the data will already be absorbed into models that are shaping our digital future. The only question left will be whether we were simply spectators to our own intellectual expropriation or whether we’ll find ways to take back some control over the most valuable thing we produce: not our goods and services, but our knowledge itself.
