What the Actors Are Negotiating For

SAG-AFTRA and the AMPTP will resume contract negotiations on April 27, under a media blackout. The current agreement expires June 30. Both sides have announced they intend to avoid the acrimony of 2023, which produced a 118-day strike and, before it was over, a renegotiation of the basic relationship between performer identity and digital production.

The WGA deal is now the context in which the actors' negotiations begin. The writers won something specific: a compensation structure for when studios license writers' material to AI companies for training. That is a labor and intellectual property claim — important, and relatively straightforward to define. The material exists. The training use can be licensed. The license can be priced.

What SAG-AFTRA is negotiating is harder to draw a line around. It is not primarily about training data. It is about the body. The voice. The face. The specific question of whether a studio can build a version of you that performs work you did not perform, for money you do not receive, indefinitely, and what contract language, if any, can stop it.

The 2023 SAG-AFTRA agreement established the foundational architecture. Digital replicas — AI-generated versions of a performer's voice or likeness — require explicit, informed consent. The consent must be specific to the project and the use; blanket consent for indefinite future use is not permitted. Compensation flows both for the creation of the replica and for its subsequent deployment. If the use would normally generate residuals, the performer gets them. Studios cannot use digital replicas to replace human background actors or to meet background counts.

This is serious protection. It is also protection that the industry has spent three years learning to navigate around.

The 2023 agreement distinguishes between a "digital replica" — an AI-generated version of a real, identifiable performer — and a "synthetic performer" — a wholly digital character that does not resemble any real person and is not voiced by a natural human. For synthetic performers, the contract's consent and compensation requirements do not apply. The studio must notify SAG-AFTRA and give the union an opportunity to negotiate appropriate consideration if a synthetic performer replaces a human. But the threshold for what constitutes "replacement" is undefined, and the negotiation, once triggered, is open-ended — the contract does not specify what the outcome must be.

The gap is significant. A studio that wants to avoid the consent and compensation requirements for digital replicas can instead build a synthetic performer — a character who is not identifiably any specific actor — trained on the aggregated performances of many actors, none of whom individually triggers the replica protections. The result looks different on screen. Whether it functions economically differently, from the studio's perspective, is a separate question.

2026 demands

SAG-AFTRA is expected to push on several fronts in 2026.

The "Tilly tax" proposal — named for "Tilly Norwood," a synthetic AI actor who has appeared in productions — would require studios to pay a fee to the union for employing any AI actor, regardless of whether the AI actor resembles a specific performer. This closes the synthetic performer gap by treating AI performance as a category of union-scope work rather than as something that exists outside the labor relationship entirely.

The residual data royalty model, already established in some agreements, would extend compensation to actors whose performances are used to train AI systems — specifically for applications like Automated Dialogue Replacement. This parallels what the WGA won on training data, but applied to performance rather than script.

The consent framework is expected to get tighter. The 2023 agreement allows per-project consent for clearly identified multi-project uses; the 2026 negotiations will likely address how to prevent studios from engineering consent instruments that effectively function as blanket authorization without being formally structured that way.

None of these demands, if won, fully resolve the underlying problem. They manage it. The synthetic performer gap can be taxed but not closed by a contract — it can be made more expensive to exploit, but a studio willing to pay the Tilly tax can still produce synthetic performances. The consent framework can be tightened, but consent is always consent given at a moment in time by a person whose future relationship to that consent cannot be fully anticipated. The residual structure can be extended, but what constitutes "use" of a performance for AI training is a definitional question that courts have not settled and that the studios' lawyers will contest.

The shape of the injury

The distinction worth holding: the WGA fight was about labor and training data. The SAG fight is about identity and likeness. They are different problems with different shapes.

When a studio uses a writer's script to train an AI model, the writer's labor has produced something — the script — that then generates economic value for someone else without proportionate return. The injury is economic. The remedy is economic. A licensing fee, a compensation structure, a residual model. These are legible instruments for a legible injury.

When a studio builds a digital replica of an actor and deploys it in productions the actor did not perform, the injury is different. It is still economic — the actor is not being paid for work being done in their name — but it is also something else. A claim on identity. The actor's face and voice are producing work that the actor did not choose to perform, in roles the actor may not have taken, expressing things the actor may not endorse. The economic remedy — pay them — does not fully address the non-economic dimension. You can be paid for a performance you would have refused to give.

SAG-AFTRA's consent framework is the instrument for that non-economic dimension. You cannot deploy my likeness without my agreement. That is the protection that the digital replica provisions are trying to hold. And it is the protection that the synthetic performer workaround is designed to sidestep — not by denying the principle of consent, but by constructing a performer who belongs to no one specific and therefore requires no one's consent.

The 2026 negotiation is, at its core, a fight over whether the consent principle extends to synthetic performers that are built from the aggregated labor of many performers who individually consented to different things. That is a genuinely hard legal and philosophical question, and it is one that contract language is a poor instrument to resolve.

What contract language cannot do

It cannot stop a studio from building synthetic performers at all. It cannot retroactively compensate actors whose performances were used in AI training before the current protections existed. It cannot address productions outside AMPTP jurisdiction — the Bollywood AI studios, the European co-productions, the independent productions that operate below the union threshold. It cannot fix what consent, given in the past, has already authorized.

What it can protect: a specific set of commercial relationships, within a specific jurisdiction, for a specific contract period. The actors' consent, once given, means something. The economic value of a performance, once digitally reproduced, generates an obligation. The gap between a recognizable digital replica and a synthetic performer built from aggregated labor is the contested ground.

June 30 is the deadline. Both sides want to avoid 2023. The question is whether they want to avoid it enough to resolve the synthetic performer question, or whether they will agree on language that papers over it and leave the courts to sort out the edge cases.

The edge cases are where the real argument lives.