New SAG-AFTRA Deal: What the 2026 AI Protections Mean for Creators

SAG-AFTRA's 2026 AI framework is the most consequential update to performer rights since the union's 2023 contract first tackled generative AI. Studios no longer get to experiment in a gray zone. Consent, compensation, and disclosure are now hard requirements, and the scope now explicitly covers voice cloning, full synthetic performers, and downstream uses that weren't even on the table two years ago.

If you're a union performer, a non-union creator, or a business licensing talent, the new rules change what you can do with AI and what you owe the humans behind the performance.

From the 2023 Strike to the 2026 Framework

The 2023 SAG-AFTRA strike ran 118 days, the longest actors' strike in Hollywood history. AI was not the only issue, but it became the defining one. The tentative agreement reached on November 8, 2023, and ratified on December 5, 2023, with roughly 78% approval, introduced the first real guardrails: a distinction between "employment-based digital replicas" and "independently created digital replicas," plus consent language for scanning performers on set.

Those guardrails were a starting point, not a finish line. Between 2024 and 2025, three things exposed the gaps:

• Voice cloning tools matured faster than the contract assumed. Commercial voice models could replicate a performer from a few minutes of audio, not the hours the 2023 language implied.
• Generative video models like Sora 2 raised the stakes for full-body synthetic performances, extending the problem beyond face scans.
• Non-signatory productions and ad agencies were using union-adjacent talent or training data in ways the contract didn't reach.

The 2026 framework addresses each of those gaps directly and tightens the definitions used in 2023.

What the 2026 Protections Actually Require

At a high level, the 2026 rules rest on three pillars: informed consent, specific compensation, and use-case transparency. Each pillar has teeth the 2023 deal lacked.

1. Informed, Specific Consent

Consent is no longer a broad sign-off buried in a call sheet. Under the 2026 framework, producers must obtain separate, specific consent for:

• Creation of a digital replica (scanning, voice capture, or model training)
• Each distinct use of that replica outside the original production
• Any alteration that materially changes the performer's likeness, voice, or performance
• Training of any AI model using the performer's captured data

Blanket "perpetual, universal" consent clauses are effectively off the table for covered work. Consent tied to a specific project cannot be silently rolled into a different project later.

2. Compensation Tied to Use, Not Just Capture

The 2023 contract required payment for scanning performers. The 2026 framework goes further, separating compensation into capture, each subsequent use, and any use that extends the performer's work into new media (for example, generating additional scenes after wrap). Residual structures mirror traditional on-camera pay rather than one-time buyouts, which was the industry's preferred model for AI-generated content.

Background actors receive specific protections here. In 2023, the union fought to ensure a day's scan couldn't yield a lifetime of synthetic extras. The 2026 rules formalize that by requiring ongoing compensation whenever a scanned background performer's likeness appears in new material.

3. Disclosure and Recordkeeping

Producers must now maintain records of:

• Which performers' data was used in any AI-generated output
• The nature of each use
• The chain of consent for that use

These records can be audited by the union. For performers, this is the difference between suspecting their likeness was used and being able to prove it.

Voice Cloning Gets Its Own Section

Voice is where the 2026 deal breaks the most new ground. Key provisions:

• Voice models count as digital replicas. Training an AI on a performer's voice requires the same consent and compensation as a face scan.
• Style is different from identity. Generic "warm female narrator" direction is allowed. A model trained to specifically sound like an identifiable performer is not, absent consent.
• Post-mortem voice use requires estate consent and is subject to the same compensation framework.
• Dubbing and localization using cloned performer voices triggers the compensation rules, closing a loophole that let some studios argue dubbing was a technical, not creative, use.

Voice actors, audiobook narrators, and commercial voiceover talent gain the clearest benefit from this section. Their work has been among the most exposed to AI substitution and the least protected by pre-2026 contract language.

Synthetic Performers and "No Human" Output

Fully synthetic performers, characters generated with no single identifiable human source, sit in a more complicated spot. The 2026 framework acknowledges they exist but requires producers to disclose when a principal role is performed by a synthetic character and to notify the union before such a character replaces a role that would otherwise be covered. It does not ban synthetic performers outright. It does make their use visible, which is the first step to bargaining over them in future contracts.

What This Means for Non-Union Creators and Businesses

Even if you don't work under a SAG-AFTRA contract, the 2026 framework sets the market standard. Three practical effects for creators and businesses:

1. Licensing voice and likeness is getting more expensive. If you license real performers for AI-assisted projects, expect contracts to mirror the union structure: specific consent, per-use fees, and disclosure.

2. Training data provenance matters. Using voice or likeness models scraped from public sources carries growing legal risk, especially where right-of-publicity laws apply (Tennessee's ELVIS Act, California's AB 1836, New York's Section 50-f).

3. Disclosure is becoming the default. If union productions must disclose AI-generated performances, platforms and advertisers will increasingly require the same from non-union work.

For coverage of how these rules intersect with copyright law, see our breakdown of who owns AI-generated art and the 2026 AI art copyright guide.

What the 2026 Framework Does Not Do

A few honest limits worth naming:

• It does not prevent studios from using AI for pre-visualization, storyboarding, or other non-performance tasks.
• It does not cover non-signatory productions, including many commercials, video games outside the Interactive Media Agreement, and most independent AI-native projects.
• It does not give performers ownership of models trained on their data. Compensation, yes. Ownership, no.
• It does not stop synthetic performers from existing or from being used in non-principal roles.

These gaps are where the next round of negotiation, likely in 2027-2028, will focus.

Key Takeaways

• Consent under the 2026 framework must be specific, not blanket, and separate consent is required for creation, use, alteration, and training.
• Compensation is tied to each use of a performer's digital replica, not just the initial capture, and background actors are explicitly covered.
• Voice cloning now gets the same treatment as face scans, closing the biggest gap in the 2023 deal.
• Disclosure and recordkeeping are auditable, giving performers a real path to verify misuse.
• Non-union creators and businesses should treat the 2026 framework as the emerging market standard for any project involving real performers.

Related Resources

• Sora 2 and Voice Cloning: The Next Wave of AI Copyright Battles
• Who Owns AI-Generated Art?
• Can You Copyright AI Art? The Complete 2026 Guide

This article is for informational purposes only and does not constitute legal advice. Contract terms vary by production and agreement type. Performers with specific questions should contact SAG-AFTRA directly or consult a qualified entertainment attorney.