The Writers' Room After the Strike

The Writers Guild of America’s 148-day strike in 2023 produced a contract that was, at the time, celebrated as a meaningful protection for human writers against AI displacement. The key provisions: studios could not use AI to write or rewrite literary material, could not require writers to use AI tools, and could not use AI-generated material as source material that would reduce a writer’s compensation. Residuals would apply to AI-generated content when it replaced human-written content. A writer’s ability to claim sole credit would not be affected by whether AI tools were used.

The provisions were real. They were also written for a technology that was static, for an industry that was cooperative, and for enforcement mechanisms that relied on the studios’ good faith.

None of those conditions held.

What the Contract Covered and What It Didn’t

The WGA provisions addressed the most obvious deployment scenarios: a studio taking a human writer’s draft and using an AI to do rewrites, cutting writing fees. That specific scenario is now genuinely constrained — the union has filed grievances and won several arbitrations on exactly this point.

The scenarios that the contract did not address, or addressed inadequately, have proliferated.

The “development AI” scenario: a studio uses AI to generate extensive story treatments, plot outlines, character analyses, and thematic explorations during the development phase, before any writer is hired. A human writer is then contracted not to develop a project from scratch but to “execute” a pre-developed AI-outlined project. The studio argues this is not using AI to write literary material — it is using AI in development, which precedes literary material. The WGA disputes this characterization. The dispute has not been resolved clearly, and several such contracts have been quietly proceeding.

The “room ratio” scenario: under pressure to produce more content for streaming platforms, several studios have reduced writers’ room sizes and substituted AI-assisted plotting for the “breaking story” function that large rooms previously performed. The human writers who remain do the dialogue writing — the part of the job that AI still does worse than competent humans — while the structural story architecture is AI-generated. The studio can comply with the letter of the WGA agreement (no AI writing) while substantially changing the ratio of human-to-AI contribution in the actual creative work.

The international production scenario: productions that originate outside WGA jurisdiction — primarily in the UK and Canada — have substantially fewer restrictions on AI use, and several US studios have quietly shifted development work to non-WGA productions.

The Voice Actor Parallel

The 2023 SAG-AFTRA strikes included parallel provisions about AI voice replication — studios could not replicate a performer’s voice or likeness using AI without their consent and appropriate compensation.

The implementation has been contested in predictable ways. The definition of “replication” is contested: if an AI voice model is trained on a performer’s vocal patterns and produces new dialogue in their style, is that replication or inspiration? Several minor characters in video games and animated series have had their voices “covered” by AI voices that are similar but not identical to the original performers — close enough to provide continuity, different enough to avoid clear technical violation.

The performer consent provision has worked better for established performers with brand recognition. For session voice actors — the vast number of less-recognizable performers who voice secondary characters, video game NPCs, commercial voiceovers — the consent provisions were negotiated with weaker specificity, and the market has contracted significantly. Several video game studios now use AI-generated voices for secondary characters entirely, citing the difficulty of managing consent and compensation for the large numbers of voice actors that complex games require.

What the Union Actually Controls

The WGA’s leverage is real but limited to the territory of its jurisdiction: writers covered by the WGA Minimum Basic Agreement, working on productions bound by that agreement, whose violations can be grieved through the established arbitration process.

This covers a substantial but not complete territory. It covers most major theatrical and streaming productions for US studios. It does not cover the rapidly growing non-union production sector, international productions, video game writing (which is covered by a separate IATSE agreement with different AI provisions), advertising and branded content, or the vast territory of online video and podcast content creation that employs writers but not under guild agreements.

The studios that are most aggressively experimenting with AI writing tools are, predictably, those at the edges of guild jurisdiction or those with the resources to fight — and win — arbitrations about contract interpretation. The studios that have been most careful about WGA compliance are the ones who most need to maintain labor relations, which is not the studios most willing to push the boundaries.

The Economics of Enforcement

There is a fundamental tension in how labor law enforcement works that advantages studios in the long run.

The WGA files a grievance when a specific violation is identified. The grievance goes to arbitration, which takes months. During arbitration, the production continues. If the union wins, the studio pays back wages and potentially modifies the practice. The cost of an arbitration loss is a fraction of the savings generated by the AI deployment during the period before the grievance was resolved. The rational studio behavior, if it believes the practice will eventually be found to violate the contract, is to deploy it anyway and pay the arbitration losses as a cost of business.

This is not unique to AI. It is how labor law enforcement works generally. The difference is speed: the rate at which AI capabilities are developing means that specific AI deployments become obsolete and are replaced by new ones faster than the grievance and arbitration process can adjudicate them. By the time a specific practice is definitively ruled a violation, the studios have moved on to a different practice that requires a new grievance cycle.

The WGA struck, negotiated real protections, and won a meaningful victory in 2023. That victory established a floor. It did not establish a ceiling. The studios are finding the space between the floor and the ceiling, and the space is large. The next round of negotiations — presumably in 2026 — will try to close some of that space. The studios will find more space. This is the dynamic of labor relations with technology, and it does not end.