Hollywood creatives are paid to train the AI that may replace them
Award-winning writers, directors and producers are quietly taking gig work that trains AI models to do their own jobs, according to a Guardian investigation published today. It's the sharpest portrait yet of the invisible hiring boom behind frontier labs — and the people powering it are often the ones most exposed to the fallout.
Hollywood pros are being paid $12 to $200 an hour to teach AI models the skills that used to pay their bills. Screenwriters, directors and producers — squeezed by a brutal industry slump — are signing on with training agencies that hold contracts with the likes of Anthropic and OpenAI to teach models the intricacies of their craft: writing a screenplay, devising a shooting schedule, assembling a pitch deck. Ruth Fowler, a screenwriter and producer whose credits include the BBC One drama Rules of the Game, told the Guardian she took the work "because I was thinking: wow, I'm always broke," describing production as "down by 35% or something insane." One anonymous LA documentary director put the trade-off bluntly: he was "essentially handed a shovel and asked to dig the grave of my profession," yet saw no point in abstaining from work that is "simply a thing that will be with us in the future."
The economics make the paradox hard to escape. Shoot days in Los Angeles fell 48% between 2021 and 2025, and US jobs in motion picture and sound recording tumbled from about 450,000 in July 2022 to 326,000 by May 2026 — a 28% drop, per Bureau of Labor Statistics data. Meanwhile the demand for human AI trainers is booming, with agencies like Mercor, Micro1 and Handshake recruiting experienced talent; Micro1 this week was seeking producers with "demonstrated mastery in film, television, digital, or live event production" at up to $85 an hour to design evaluation tasks simulating real production scenarios. Netflix this month revealed it used AI in 300 of its 1,000 titles this year, and directors including Ron Howard are embracing the technology rather than fighting it.
The quiet tragedy is that the workers training the models are the ones with the clearest read on what the machines still can't do — and that they're often glad of the money anyway. A TV writer with credits across Amazon, HBO and NBC said AI "no closer to approximating human emotion," capable of a procedural like CSI but not Severance or Succession. Jody Wheeler, a USC lecturer and screenwriter, was more sanguine: the shovel can bury a profession, he argued, or help people "unearth stuff that they wouldn't have been able to do themselves before." Take it as a reminder that behind every headline about model capability sits a workforce quietly subsidizing the disruption aimed at their own careers.
What to watch: whether this training-gig economy becomes a regulated labor market of its own, and whether the people building it get any seat at the table when the models arrive.
If you could see what a model was trained on, would you still work for the lab that's building it? Tell us in the comments.
