THE LINKEDIN MESSAGE was polite, professional, and — to anyone who read it carefully — breathtaking in its audacity. Mercor, a San Francisco-based startup valued at $10 billion, was reaching out to visual-effects artists who had spent decades building blockbuster worlds for Disney and Sony, offering to purchase their prior production work: 4D physics scenes, depth maps, motion-tracking data. The kind of high-end material that trains AI models to do, eventually, what those same artists do for a living. Several artists who received the pitch pointed out an inconvenient detail: they do not own the work, their former employers do, and selling it would violate the NDAs that govern virtually every studio contract in Hollywood.
Mercor insists it only licenses content from individuals who own it, and requires signatories to attest as much — though it does not verify ownership. The company's outreach, first reported by the Wall Street Journal, nonetheless illuminates a deeper structural tension in the AI economy. Frontier labs have an insatiable appetite for specialized, domain-expert training data; the easy internet corpus has largely been consumed. Epoch AI, a research institute, projects that the stock of high-quality public text data will be fully utilized between 2026 and 2032. As that wall approaches, the hunt for fresh material has shifted from scraping the open web to soliciting the proprietary work of professionals — and the professionals in question are, increasingly, the very people whose livelihoods AI threatens to displace.
Rendering unto Caesar
Yet the VFX industry's predicament extends well beyond one startup's ham-fisted outreach. A study cited by the Hollywood Reporter found that roughly a third of entertainment executives expected more than 20 percent of industry jobs — an estimated 118,500 positions — to be eliminated by 2026, with visual effects and postproduction flagged as especially vulnerable. Eighty percent of early AI adopters in entertainment already deploy the technology in postproduction workflows. The artists being asked to sell their work are not doing so from a position of strength; many are freelancers who weathered COVID shutdowns, the 2023 writers' and actors' strikes, and a broader contraction in studio spending. One VFX artist who spoke to the Journal captured the dilemma neatly: some colleagues refuse Mercor's overtures on principle, while others, out of work and out of options, simply do not care.
This is the flywheel that makes the Mercor episode more than a footnote about IP hygiene. The industries most vulnerable to AI-driven displacement are also the richest repositories of the specialized data that frontier models need to improve — and the economic pressure created by displacement makes workers in those industries more willing to sell. Mercor's own trajectory illustrates the scale of demand: the company quintupled its valuation from $2 billion to $10 billion in eight months, now manages over 30,000 contractors paid a collective $1.5 million per day, and counts OpenAI, Anthropic, and Meta among its clients. Its rise accelerated after Meta's $14.3 billion investment in rival Scale AI prompted several major labs to seek alternative data vendors. The training data supply chain, hitherto dominated by internet scraping and crowd-sourced labeling, is rapidly professionalizing — and the professionals it needs are increasingly desperate.
The legal exposure, meanwhile, cuts in multiple directions. Anthropic recently agreed to pay at least $1.5 billion to settle a copyright-infringement lawsuit over the use of pirated books in training data. A company that knowingly approaches workers bound by confidentiality agreements or copyright assignments could face its own liability, according to attorneys involved in AI IP litigation. And Mercor itself has fresh reasons to worry about the security of whatever data it does acquire: in late March, the company confirmed a supply-chain cyberattack that reportedly exposed four terabytes of data, including material linked to its AI lab clients. A class-action lawsuit has already been filed on behalf of more than 40,000 affected contractors and customers. Meta has reportedly halted its partnership with the company pending investigation.
The deeper question is whether the current model — intermediaries cold-messaging freelancers on LinkedIn to buy work they may not own, for use by labs whose models may render those freelancers obsolete — is a sustainable architecture for the AI economy, or merely the awkward adolescence of a supply chain that has not yet figured out how to pay for what it consumes. One VFX artist, asked by Mercor to sell his prior studio work, offered a counter-proposal: he would create new work, at his standard rate. He has not heard back. ■
For more, join 75,000 subscribers getting tech's favorite brief.
