Hollywood Writers Secretly Training the AI That May Replace Them
Summary
- • A Hollywood showrunner now secretly works as an AI trainer to make ends meet
- • Post-2023-strike entertainment industry never recovered, forcing writers into AI gig work
- • Writers earn $52–$150/hour annotating data, red-teaming LLMs, and generating content for safety testing
- • The same writers who struck against AI are now training the systems that threaten their careers
Details
Hollywood writer turned AI trainer after post-strike industry collapse
A working TV showrunner with credits on Paramount, Hulu, and the BBC began taking AI training contracts in September 2025 after a producer defaulted on a six-figure payment. The 2023 WGA strike ended after nearly five months but failed to restore industry momentum, leaving many writers unemployed and in debt.
Unemployed Hollywood writers openly recruiting each other into AI training work
An unofficial WGA Facebook group became a clearinghouse for financial distress and AI gig leads. One writer posted that Mercor was paying $150/hour for writers, calling it 'easy money.' The group was filled with posts about debt, food banks, and part-time job searches.
AI training tasks span annotation, safety red-teaming, and harmful content generation
Work includes evaluating chatbot tone, annotating images and video with precise timestamps, and red-teaming LLMs by attempting to elicit bomb-making instructions, political violence incitement, and other harmful outputs. These are structured safety tests designed to probe model weaknesses before deployment.
Writer hired at $52/hour as a 'generalist' annotator after 20 unpaid test hours
Getting hired required filling out 10 applications, completing roughly 20 hours of unpaid screening tasks, and being interviewed by an AI recruiter agent. The $52/hour 'generalist' rate sits below an 'expert' tier. Other platforms reportedly pay up to $150/hour for writers.
Article argues the same writers who struck against AI are now training it to survive
The author frames this as a core irony: the 2023 strike was explicitly aimed at preventing AI from replacing writers and actors. The post-strike recovery never materialized, and financial desperation has driven those same workers to supply the labor that makes AI systems more capable and safer to deploy at scale.
AI platforms use anonymized worker identities across different clients
Workers are identified by alphanumeric codes rather than names, reflecting the commoditized, anonymous nature of data annotation labor. The author operates under different identifiers across platforms.
Real user prompts fed to annotators reveal emotional dependency on AI chatbots
The author describes prompts from actual users as 'quirky and sad and heartbreaking' — people asking AI whether their feelings are valid or whether they are lovable. This underscores the human stakes on both sides of AI development: the workers training the models and the users emotionally relying on them.
Context = background/framing, Industry Update = sector-level shift, Tech Info = how the work functions, Financials = pay/hiring economics, Insight = attributed analysis or argument, Other = notable detail
What This Means
The post-2023-strike collapse of the Hollywood labor market has created a pipeline of highly skilled creative workers into AI training gig platforms — meaning the industry's most vocal opponents of AI replacement are now directly contributing to the capability and safety of the systems they feared. This dynamic is not unique to entertainment: it reflects a broader pattern where economic pressure converts displaced workers into AI labor suppliers, accelerating the very automation that displaced them. For the AI industry, access to professional writers as annotators and red-teamers likely improves model output quality in ways that commodity labor cannot match. For the workers, it is financially necessary but structurally self-undermining.