Contract case v0.3 · Prepared
Performers: whose work, whose double, whose income?
Human by choice · One human choice
Imagine an actor offered a day's work and asked for permission to create a digital double. The double could later appear in a scene the actor never performs. A different production might use a human-looking character, voiced by no person, with no identifiable performer copied or employed in that role. These are invented situations. They raise different questions about work, identity and who receives the money.
Keeping work human could mean requiring a person to perform, letting a person control a digital copy, or sharing the gains when a production uses a substitute. Agreeing with one of those aims does not settle the others.
The 2026 SAG-AFTRA television and film agreement, ratified on 4 June and effective from 1 July 2026, makes those differences concrete. This case examines its published agreement and explanations, rather than all entertainment contracts, later changes or UK legal rights. It is not advice about an individual's contract.
1. Which task would a person still perform?
Separate performing a scene from permitting a digital depiction and bargaining over its use. A person can retain control or income without doing every scene in person. Conversely, a credit or payment does not establish that a person actually performed the scene.
A covered synthetic must look human without being recognisable as a particular performer; the category excludes human voicing, digital replicas and employment in that role. SAG-AFTRA’s explanation describes a preference for human performances. This defined category concerns covered work, rather than every computer-generated character.
Define the use before choosing the rule. Is it a new performance, an alteration of recorded work, a recognisable copy, or a different character? An audience label saying “made with AI” would not answer those questions or determine anyone's pay.
2. What are we actually protecting?
Its page 132 footnote allows digital replicas to count as human performance here. That preference therefore need not preserve in-person working days.
That distinction need not make the protection worthless. Control over one's identity and a share of income can matter even when software supplies an image or voice. But they protect something different from a working relationship, the opportunity to practise a craft or the experience of watching a live performance.
A proposal should name its purpose. If it promises jobs, count opportunities to work. If it promises control, examine the permission and the ability to refuse. If it promises a share of AI wealth, follow the income. A single “human involvement” label cannot demonstrate all three.
3. Does consent give a person a real choice?
Consider the invented actor again. A specific, understandable permission request would be more informative than a broad reference to future technology. Yet the person might still fear losing the job if they refuse. A signed form alone would not tell us whether that fear shaped the decision.
These are questions for evaluating an arrangement, not findings that this agreement has failed or that all performers face the same pressure. Ask when permission is sought, what uses it covers, what can be negotiated, and what happens after refusal. Establish who can obtain advice and challenge an unauthorised use. Do not assume that a familiar face implies permission, or that permission for one use settles every later one.
Replica rules have exceptions; check the relevant provision before assuming every edit needs consent or extra payment.
4. Who gains, who pays and who gets left out?
Permission to earn from a replica may benefit the person copied while leaving fewer opportunities for someone who might otherwise have been hired. An established star and a new entrant could have different interests. These are possible distributional effects to investigate, not measured consequences of the agreement.
For covered synthetics, §64.1.A makes the producer initially assess significant additional value, then notify and bargain. If talks fail and use proceeds, the union may seek monetary arbitration. Any payment goes to the relevant health or pension/retirement plan at the union’s election; none is guaranteed. This is no general veto.
This offers a collective response to replacement alongside individual control of a replica. It does not establish the amount paid or opportunities preserved. Who can actually qualify for a pooled benefit, including people seeking their first work? That needs checking; this case does not assert the schemes’ eligibility rules. A fund contribution and a working day remain different outcomes.
5. Who can make and enforce the choice?
The union and producers negotiated these limits. Exceptions cover customary digital depictions of activities a human cannot perform without serious risk to life or health, and non-human characters created without covered-performer services. Name-and-feature prompting that creates recognisable principal facial features requires individual consent and bargaining, with constitutionally protected-use exceptions. Different routes can give different people power to enforce a protection.
UK performers already have their own organised agenda. In a 25 September 2026 statement, Equity described advanced talks with Pact over transparency, consent and remuneration, alongside its campaign for wider personality rights. That dated statement is not verification of the negotiations' subsequent outcome or a claim that new UK rights are in force.
A useful contribution from this programme is to make the choices understandable across sectors, while recognising the people already doing the bargaining. For a proposed arrangement, the parties could specify covered uses, notices, information access, payment, exceptions and remedies together. That is an agenda for discussion, not an adopted model clause or a substitute for workers' representatives and jurisdiction-specific legal assessment.
6. What would change our judgment?
Imagine a small production that could afford a digital character but not an additional shoot. Compare human delivery, a paid replica, a covered synthetic and no production. The last option would also remove the proposed work for other people and the experience offered to audiences. This is an invented comparison, not evidence that substitutions create net jobs. It challenges the assumption that a ready-made human job always exists to be protected.
Digital substitution might also enable creative choices or avoid dangerous work. Safety is a serious general argument; the narrow exception above already addresses it. The question is whether a particular restriction prevents valuable activity, displaces existing work, or channels savings into higher margins. None of those outcomes follows from the technology’s label.
Compare production costs, opportunities and earnings across established and newer performers, and who receives any savings. Examine whether people understand permissions, can refuse without retaliation, discover later uses and obtain effective remedies. Include the cost of advice, records, monitoring and dispute resolution; rights that only well-resourced people can enforce may distribute gains unevenly.
If an arrangement expands creative work while giving people effective control and a fair share, that would support it. If it concentrates income while eroding entry routes or making refusal impractical, stronger limits or different bargaining terms may be justified. These are proposed evaluation criteria, not observed results or a validated research protocol.
Before supporting a “human” rule, ask what it retains: a person doing the work, a person controlling its use, or a person sharing the proceeds. Which of those matters here—and whose choice is it?
Human by choice is Policy Window's advocacy programme on AI, work and who decides. Its analysis and proposals are distinct from the neutral Policy Window wiki. This case explains existing union work; it claims neither an original bargaining mechanism nor a representative mandate.