SAG-AFTRA's synthetic likeness deal left the most valuable rights on the table

The past two years produced a wave of high-profile agreements between Hollywood studios and talent unions promising to govern how AI can replicate human performances. SAG-AFTRA's interim agreements in 2023 and 2024, WGA's hard-won AI provisions, and individual deals signed by studios with synthetic media vendors were presented as proof that the industry had gotten ahead of the problem. Inside the trades, the settlement framing stuck. Synthetic likenesses: solved, or at least negotiated.

That framing is wrong in ways that will cost studios, distributors, and talent alike.

The consensus view: contracts contain the risk

The consensus argument runs as follows. Unions secured consent and compensation requirements before studios could industrialise synthetic performance at scale. SAG-AFTRA's AI provisions require a performer's informed consent before their digital likeness can be used and mandate that synthetic replicas be paid at rates comparable to the original performance. Studios, for their part, argue they have internal governance processes: legal sign-off on likeness acquisition, IP audits on AI vendor contracts, and indemnification clauses in production agreements. The view from legal departments in Burbank and Century City is that deepfake talent is an HR and clearance problem, no different structurally from music sync or location rights, and one that experienced rights teams can manage with the right paper trail.

This position has genuine merit. The contracts that emerged from the 2023 strikes do contain sharper language than anything that existed before. Studios that ignored likeness consent entirely, as some digital production houses were doing with historical footage of deceased performers, now face explicit contractual liability. That is real progress.

Where the consensus breaks down

The gap the deals don't close is not about consent for primary use. It is about what happens downstream, across three dimensions that the current frameworks barely touch.

First, residual value and secondary markets. When a performer's likeness is captured once under a union contract, the synthetic version can theoretically travel indefinitely into derivative content: localised versions for international markets, promotional materials outside the scope of the original production, AI-generated cameos in content produced years later. Residual structures in Hollywood were built around reruns and home video windows, categories with defined durations and distribution channels. Synthetic likeness use does not mapmapUsing software to automate repetitive marketing tasks and campaigns, enabling personalisation at scale across channels like email, web, and social.View full definition → onto those windows. A digital replica of an actor can appear in a Spanish-language dub, a social media campaign, or an in-game promotional spot, each with its own revenue stream, none of which existing residual schedules cleanly address. The underlying economics of how talent contracts were designed assume scarcity of performance; synthetic replication removes that scarcity, and the pricing frameworks have not caught up.

Second, personality rights across jurisdictions. The US has no federal right of publicity. State laws vary from California's robust post-mortem protections under Civil Code 3344.1 to states with no statute at all. Meanwhile, the EU's GDPRGDPREU regulation governing how organizations collect, store and use personal data, with fines tied to global revenue for breaches.View full definition → creates a different but overlapping framework through biometric data classification, and several Southeast Asian markets where streaming revenue is growing fastest have almost no personality rights law to speak of. A studio that obtains clean consent under California law for a synthetic replica can find that same replica used in ways that would be tortious elsewhere, by a licensee the studio no longer controls. Enforcement across these jurisdictions is not a paperwork problem; it is practically and financially prohibitive for individual performers and often for mid-tier studios as well.

Third, and most overlooked: the vendor layer. Most production companies are not building their own synthesis models. They are licensing tools from third-party vendors, some of them well-capitalised and some not, whose training data provenancedata provenanceData lineage maps how data moves and transforms across systems, from origin to consumption, showing where it came from, what changed it, and where it goes.View full definition → is opaque. When a studio licenses a synthetic voice that turns out to have been trained partly on a performer's archived recordings without consent, the indemnification clause in the vendor contract is not a solution. It is the beginning of litigation. The chain of accountability for AI-generated content involving real likenesses runs through layers of subcontracting that studio legal teams have not yet mapped with the same rigour they apply to music clearance.

There is also a structural incentive problem that the industry is reluctant to name. Synthetic likenesses of established talent are most valuable to studios because they reduce production cost, particularly for marketing, reshoots, and international versioning. The performers with the most leverage to negotiate protections, A-list talent with strong agency representation, are already protected by individual deal riders that go beyond union minimums. The performers with the least leverage, day players, background artists, performers whose likeness was captured in content produced before any AI provisions existed, are exposed in proportion to their inability to negotiate. This is not unique to entertainment, but the specific economics of how studios extract margin from catalogue content makes it acute here.

What a sharp operator should actually do

The first move is not to wait for federal right-of-publicity legislation, which has moved slowly through Congress for years and shows no sign of accelerating in 2026. Treat international personality rights as a supply chain audit item on every co-production and licensing deal. Before any synthetic likeness is cleared for use in a non-US market, require the same jurisdictional sign-off you would require for a music sync cleared in one territory but exploited globally.

The second move is to build vendor diligence into production budgets rather than treating it as a legal afterthought. Require model training data audits from any synthesis vendor, with representations about consent and provenance that carry real indemnification weight, not boilerplate. Several studios have begun requesting these as of mid-2026; the ones that do not will be the defendants in the cases that set the next round of precedents.

The third move is to map your catalogue exposure now. Any production involving living performers made in the last fifteen years likely contains biometric data that could be extracted and used to train synthesis models. Understanding what you hold, and under what terms, is defensive work that also has commercial value: studios that can demonstrate clean provenance on their synthetic assets will have an advantage in the licensing markets that are forming around authenticated AI content.

The rights crisis around synthetic likenesses is real, but it is not primarily a crisis of bad actors. It is a crisis of contractual frameworks designed for one production model being applied to a completely different one. Studios that treat their current agreements as a solved problem are carrying exposure they have not priced.

The full course on this sector:AI in Media & Entertainment.