Hollywood’s AI fight shifts from adoption to control of labor, likeness and rights

The central economic feature of generative AI is not simply that it can produce text, images, voices or moving pictures. It is that creative material can become a reusable input: a performance can become a digital replica, an archive can become training data, and a generated draft can enter a workflow that previously began with paid human labor.

That changes bargaining because the party controlling the reusable asset may capture value across multiple stages or projects. Studios and technology providers could benefit from faster iteration, lower marginal production costs and the ability to modify or localize material without recreating every underlying element. Established performers and creators may also benefit if they can license valuable likenesses, voices, catalogues or specialized styles on favorable terms.

The risk falls most heavily on workers whose contributions are easier to fragment into tasks. AI does not need to replace a screenwriter, performer or production artist completely to change employment. If it performs preliminary drafting, background generation, temporary voice work or some forms of alteration, the number, duration or price of human assignments could decline. Entry-level roles may be especially exposed because they often involve the repeatable work through which people acquire experience and professional relationships.

Contracts turn technical capability into a rights question

Hollywood’s labor agreements increasingly intervene at the point where a technical capability becomes a commercial use. The WGA framework preserves the covered writer’s status by declaring that AI is not a writer and that generated material cannot be used to downgrade a writer’s compensation or credit position. It also allows a writer to choose AI with company consent, making the arrangement a regulated option rather than a categorical prohibition.

SAG-AFTRA’s 2026 agreement applies a related mechanism to performance. Its protections cover replicas derived from ordinary production photography as well as dedicated scans, reducing the possibility that a producer could escape contractual duties because no formal scanning session occurred. The agreement also requires an articulable business reason when seeking consent to scan, adds security and transfer protections, and establishes a stronger bargaining process around synthetic human performances.

These provisions do not demonstrate that AI has preserved or eliminated jobs. They instead redistribute decision rights. A producer’s ability to generate an asset does not automatically settle whether it may be used, for how long, in which production or for what payment. Consent, documentation and enforceability become part of the production infrastructure.

Copyright creates a second layer. The Copyright Office’s position that human-authored elements may be protected, while prompts alone generally do not confer authorship, gives companies an incentive to preserve evidence of meaningful human creative control. Its training report also rejects a single answer for every dataset or model use, distinguishing potentially transformative applications from commercial systems whose outputs compete with the works used to train them.

The likely near-term result is therefore uneven adoption rather than wholesale automation. Uses with clear permissions, measurable savings and limited reputational risk should advance fastest. Uses that depend on disputed training material, ambiguous ownership or replacement of recognizable human labor face higher legal and industrial-relations costs. Hollywood’s decisive AI capability may consequently be rights management: knowing what material entered a system, who authorized it, what the output contains and who must be paid.