SAG-AFTRA 2026 TV/Theatrical Agreement: the AI and digital replica provisions
The agreement took effect on July 1, 2026 and runs through June 30, 2030. Its digital replica provisions turn consent into a documentation problem: not whether the performer agreed, but whether you can show, later, exactly what they agreed to.
At a glance
- Authority
- SAG-AFTRA and the AMPTP — General Provisions §§ 64 and 64.1
- Jurisdiction
- United States — union agreement
- Effective
- July 1, 2026
- Expires
- June 30, 2030
Who this applies to
- Productions working under the SAG-AFTRA TV/Theatrical Agreement on or after July 1, 2026
- Signatory producers scanning, de-aging, crowd-replicating or otherwise creating a digital likeness of a covered performer
- Post and VFX vendors delivering to a covered production — indirectly, through the production’s own obligations
Dates
| Date | Event |
|---|---|
| June 4, 2026 | Ratified by SAG-AFTRA membership, 91.42% in favor |
| July 1, 2026 | Agreement takes effect |
| June 30, 2030 | Agreement expires |
What changed
The 2026 Theatrical/Television Agreement between SAG-AFTRA and the AMPTP was ratified on June 4, 2026, with 91.42% of voting members in favor. It became effective on July 1, 2026 and expires on June 30, 2030.
The AI and digital replica provisions sit at General Provisions §§ 64 and 64.1. For most producers the consequential change is not that digital replicas became permissible — replicas were already being negotiated deal by deal — but that the agreement fixes a consent standard with a specific evidentiary shape. Consent must be clear and conspicuous, and it must be based on a reasonably specific description of the intended use of the Digital Replica.
Those two phrases carry the weight. “Clear and conspicuous” governs how the consent is presented: it is not satisfied by a clause a performer would not reasonably notice, folded into the middle of a long-form deal memo. “A reasonably specific description of the intended use” governs what the consent actually covers. A blanket grant of the right to use a performer’s replica in any medium now known or hereafter devised is close to the paradigm of what the standard is aimed at, because it describes no particular use at all.
The agreement also distinguishes between an Employment-Based Digital Replica — one created using a performer’s work under an employment contract, such as a scan session or plates shot on the production — and an Independently Created Digital Replica, made without the performer being employed to create it. Both require consent. The second carries an additional documentation burden, because you also have to be able to say where the underlying material came from.
Reading the consent standard the way a reviewer will
The practical test is not whether the performer signed something. It is whether a person reading the file two years later — an E&O underwriter, a network standards reviewer, a distributor’s delivery desk, or opposing counsel — can match the use that shipped to the use that was described at the time of consent.
A description like “digital replica may be used in the Picture and in any and all media” fails that test: it does not describe a use, it describes a universe. A description like “replica of Performer’s face composited onto a stunt double for the rooftop sequence in Episode 104, and for continuity fixes to dialogue already performed by Performer in Episodes 101–106, for use in the Series and its trailers and promotional materials” can be checked. Someone can hold that sentence against the finished cut and say yes or no.
That is the reason the intended-use description belongs in the workflow before the scan session, not after it. Once material exists, the description tends to get written to fit what was made, which is precisely backwards from the standard the agreement sets.
Where productions get this wrong
Four patterns account for most of the gaps. First, consent obtained for the shoot but never extended to the use invented in post — the de-age nobody planned, the line replaced with a voice model, the actor placed in a scene they were not called for. The consent instrument on file describes a scan; the finished picture contains something else.
Second, the rider is signed but the intended-use description is written afterward, sometimes by an assistant, sometimes copied from another project. Third, background and crowd replication gets treated as a technical process rather than a likeness question, because the vendor described it as a “tile” or a “population pass” rather than as replicas of identifiable performers.
Fourth, and most damaging at delivery: nobody kept a shot-level record. The production can produce signed consents, but cannot say which shots the replica actually appears in. That is the question a delivery desk asks, and an unanswerable version of it is what turns a two-week delivery into a two-month one.
How this interacts with everything else
The union agreement is a floor for covered work, not the whole picture. California, Illinois and New York impose their own informed-consent requirements for digital replica use, which apply regardless of whether a production is signatory. New York separately requires a conspicuous disclosure when a commercial advertisement depicts a synthetic performer. If your title reaches the EU, the AI Act’s Article 50 transparency obligations attach to the content itself rather than to the employment relationship.
These regimes do not neatly nest. A consent that satisfies the union standard is not automatically sufficient under a state informed-consent statute, and neither of them produces the machine-readable marking Article 50 contemplates. The practical answer is a single record built once — who consented, to what use, when, evidenced how — that can be presented to each of them in the form they expect.
What you must actually do
Identify every replica before the scan, not after
Walk the script and the VFX bid together and mark each shot or sequence where a performer’s face, body or voice will be generated, replicated, de-aged or extended. That list is what the consent has to cover.
Write the intended-use description first
Draft the reasonably specific description of intended use before the consent is presented, and have the person who will actually supervise the work check it. If the description cannot be written yet, the scope is not settled enough to ask for consent.
Paper it as a rider or a separate signed writing
Consent may sit in the employment contract only where that provision is separately signed or initialed by the performer; otherwise it must be a separate signed writing. Unlike the Commercials Contract, the TV/Theatrical agreement prescribes what consent must contain rather than a standardised form — there is no jointly drafted rider for it, so the instrument you use is your own. Keep the executed copy with the intended-use description attached to it, not referenced loosely.
Keep a shot-level clearance log
Record each replica against the shot or sequence, the performer, the consent instrument and its date. This is the artifact that answers delivery and underwriting questions without a re-audit.
Re-paper on scope change
If the use changes after consent — a new platform, a new territory, a sequel, an unplanned de-age — treat it as a new consent, not an interpretation of the old one.
Push the same standard down to vendors
Post, VFX and AI tooling vendors should attest to what they used, on what material, and confirm they did not train on performer material outside the described scope.
Where the law is still open
These are the live edges — the points where the text stops and drafting practice takes over. Here is what to watch on each, and how careful practitioners are handling it now.
- “Reasonably specific” is a standard, not a rule — no published bright line fixes how granular a description has to be. Draft to the depth a stranger could check against the finished cut. That is the level the conservative end of the bar is working to, and it is the level that survives review.
- Material captured before July 1, 2026 and re-used after it is settled deal by deal, not by the agreement. Fix the position in writing at the point of re-use rather than arguing it at delivery.
- Enforcement runs through the agreement’s own grievance and arbitration machinery, so there is no public body of determinations to calibrate against yet. Until there is, the drafting standard is set by practitioners, not by decisions — watch how the studio-side forms move.
- Where the union standard and a state informed-consent statute both apply, no authority has described how they interact when their requirements diverge. Build the consent to satisfy both; neither regime treats the other as a substitute.
Questions this entry answers
The same material, in the form it usually arrives — as something that happened on a job, before anyone knew which agreement or statute it belonged to.
- Do we need consent to de-age an actor?
- Yes, and the consent has to describe the de-aging specifically. Under the 2026 SAG-AFTRA TV/Theatrical Agreement, which took effect July 1, 2026, consent for a digital replica must be clear and conspicuous and based on a reasonably specific description of the intended use. A consent obtained for the shoot does not extend to a de-age invented in post — that is a different use, and it needs its own instrument before the work happens rather than after.
- Our contract already says we can use the performer’s likeness in all media. Isn’t that enough?
- That language is close to the paradigm of what the current consent standard is aimed at. Both SAG-AFTRA agreements require consent based on a reasonably specific description of the intended use, and a grant covering all media now known or hereafter devised describes no particular use at all. California goes further in the same direction: Cal. Lab. Code § 927 makes a replica provision vulnerable where the use is not described with particularity and the performer was not represented by counsel or a union representative. Breadth is exposure here, which inverts the usual instinct in rights drafting.
- We used AI to multiply a crowd. Is that a likeness question?
- It usually is, and it is the case productions most often miss. If the crowd was built by replicating identifiable background performers, each of those performers has a likeness in the finished picture and the replica consent standard reaches them. If the figures were generated rather than replicated, no individual’s likeness is involved — but they may be synthetic performers, which raises the New York disclosure obligation in advertising and, under California’s substitution test, the question of whether they replaced background performers who would otherwise have been hired. Vendors describe this work as a "tile" or a "population pass", which is why it reaches nobody who would have flagged it.
Does this apply to your production?
Answer a five-minute intake — contract type, AI usage, airing states, platforms, EU distribution — and ClearFrame returns the requirements that apply to your specific production, with citations and effective dates. Free, one project, no card.
Sources cited by the ClearFrame corpus
The sources the rules corpus records for this entry. Follow them through to the primary text.