Hollywood writers secretly train AI as entertainment industry struggles

The people training AI systems are not anonymous tech workers. Many are experienced professionals from industries decimated by the very technology they now help build. Ruth Fowler writes for Wired about her experience as a Hollywood screenwriter and showrunner who spent eight months working for AI training companies to pay her bills.

Fowler describes a sprawling gig economy in which workers are hired and fired with no notice, paid inconsistently, and managed through Slack channels and late-night messages. She worked for companies including Mercor, Outlier, Turing, Task-ify and Handshake, completing tasks ranging from chatbot evaluation to video annotation and AI safety testing, known as red-teaming.

The work began after the 2023 Hollywood strike, which was partly fought to limit AI’s role in film and television production. When the industry failed to recover, Fowler and many fellow Writers Guild of America members began searching for alternative income. AI training appeared to offer a solution. A comment in a WGA Facebook group advertised $150 per hour for writers. The reality proved far less generous.

Fowler describes an exhausting cycle of unpaid onboarding, weeks of waiting with no tasks, sudden project launches in the middle of the night, and abrupt terminations. “These are not jobs, these are tasks, and we are taskers,” a team leader told her. Workers are classified as independent contractors, which strips them of standard employment protections including stable scheduling and protection against sudden dismissal.

Pay rates have fallen sharply. In early 2025, expert-level positions advertised $150 per hour. By early 2026, many expert roles paid around $50 per hour, while entry-level annotation work dropped to $16 per hour, below California’s minimum wage for some workers. Several lawsuits have been filed alleging that Mercor misclassifies workers as contractors rather than employees.

Mercor told Fowler it employs around 300 full-time staff while maintaining approximately 30,000 independent contractors. The company said it tries to give workers “as much notice as possible when projects change.”

Fowler describes the workforce as predominantly professionals in their thirties and forties, many with advanced degrees and significant industry experience, managed by people fresh out of university. Quality control was opaque and inconsistent. Workers were scored on a scale of one to five, with little clear guidance. Promotions into reviewer roles came with no pay increase.

The psychological toll was significant. Fowler describes abandoning family meals, snapping at her child, and working through the night to secure tasks that might disappear by morning. Colleagues reported similar experiences across Reddit communities and private Discord servers.

Fowler concludes that the industry’s drive to build more capable AI systems is being powered by a precarious workforce kept deliberately unstable. “To make the machine more human,” she writes, “they will make us more like the machine.”