SAG-AFTRA AI deal: the 2% residual problem

September 21, 2026 · 5 min · Free, with full transcript

Studios and SAG-AFTRA signed likeness protections and called the AI crisis handled. Does a 2% residual for synthetic performers actually cover the exposure? The position here is no: the payment is front-loaded onto a base studios control, consent works as a one-time door, and personality rights travel with the file into jurisdictions with no equivalent of California's right of publicity. You walk away able to see where the value sits now, in training data and model weights rather than the filmed performance, and why post-mortem likeness rules ranging from zero to 70 years break estate licenses. The closing move is a contract clause that attaches residuals and consent to any model trained on your likeness, wherever it is hosted or fine-tuned.

Listen to the podcast

4 min

Chapters

Key takeaways

  • Renegotiate licenses so they attach to the model and its weights, not the single performance.
  • Demand a clause stating any model trained on your likeness carries your residual and consent wherever it is hosted.
  • Check where rendering physically happens, since a US contract gives no standing if the synthetic labor occurs offshore.
  • Treat vendor claims like OpenAI's 70% cost reduction and Hugging Face's million-model count as self-serving until independently benchmarked.
  • Assume independent producers fine-tuning open models on public footage fall outside the union agreement entirely.
Read the full transcript

Host:Leaders' Insights Five minutes on SAG-AFTRA's synthetic likeness deal left the most valuable rights on the table. Two percent. That's the residual bump SAG-AFTRA secured for synthetic performers in the 2026 studio deal. And it's the number that tells you the whole thing was a surrender.

Expert:It's worse than a surrender because two percent is calculated on a base that studios control. They license your digital double and pay you a fee up front. The residual, the money you get every time the work airs again, barely exists on the back end. They front-loaded the payment precisely so the ongoing stream shrinks to nothing.

Host:The unions came out and called this a win. Actors kept their consent rights. What did they actually leave in the room?

Expert:Consent got the headlines, but consent is a one-time door. Once you open it, personality rights, the legal claim you have over your own face and voice, travel with the file, and the deal doesn't follow them across borders. A studio can render your likeness in a jurisdiction where those rights barely register and enforcement collapses.

Host:Give me the border problem in concrete terms. Where does it break?

Expert:Take a scene generated by a model trained partly on your prior films. The rendering happens on servers in a country with no equivalent of California's right of publicity statute. You signed a U.S. contract, but the actual synthetic labor occurred somewhere your union has no standing. Who do you sue and in which court? The deal is silent.

Host:You're describing a loophole. Studios would say the tech isn't there yet to make this cheap.

Expert:The tech is closer than the contracts assume. OpenAI's own material this year claims their video models cut synthetic character generation costs by roughly 70 percent, though they sell the tooling, so treat that figure as marketing until O'Reilly Radar or someone independent benchmarks it. Even if the real number is half that, the economics of replacing a background scene or a dead actor's cameo already work.

Host:Dead actors? That's the part that makes people flinch. What does the deal say about the estate side?

Expert:Almost nothing enforceable. Most Mortem likeness is governed state by state in the U.S. 20 years of protection in some places, 70 in others, zero in a handful. The union agreement layers consent on top of that patchwork but doesn't harmonize it. So an estate signs one license and discovers the protections evaporate the moment production moves.

Host:Let me push on the money again. If residuals are gutted, where does the value actually sit now?

Expert:In the training data and the model weights, the stored parameters that are effectively the digital performer. Whoever holds those holds the asset. Hugging Face, the platform where a lot of these models get hosted, has said the number of publicly shared video generation models passed a million this year. That's their figure, and they run the repository, so it's self-serving. But even discounted, it tells you the tooling is everywhere, not locked inside three studios.

Host:So an actor's face could be sitting in a model somebody uploaded, outside any contract at all. What's the exposure nobody priced?

Expert:The union deal governs the studios it bargained with. It does not govern the independent producer who fine-tunes an open model on a decade of your public footage. Your likeness leaks through the training pipeline. And there's no residual, no consent form, no jurisdiction that reliably catches it.

Host:Studios declared the crisis managed. You're saying they managed the cheap part?

Expert:They managed the part with cameras and payroll. The expensive part, the perpetual, borderless, model-embedded version of you, got waved through because measuring it is hard and litigating it is harder. Declaring victory was the negotiating tactic.

Host:If I'm an actor or an agent listening right now, what's the one move this week?

Expert:Re-negotiate the license to attach to the model, not the performance. Demand a clause that says any model trained on your likeness carries your residual and your consent, wherever it's hosted. Whoever fine-tunes it downstream, the performance is the old asset. The weights are the new one, and right now you're signing them away for 2%.

Host:Chase the file, not the footage.

Expert:Chase the file, the footage already left the building.

Host:What we read for this one, ours Technica AI, TechCrunch AI, the decoder, KD Nuggets, hugging face, vendor, AI platform, O'Reilly radar. Done for today, there's a new AI piece every morning at MBA-training.com.

Read and go deeper