SAG-AFTRA and WGA ask judge to reject AI's fair use defense

SAG-AFTRA, the American Federation of Musicians and both Writers Guilds asked a federal judge on Thursday to reject OpenAI's fair use defenses, arguing that training a model on actors', musicians' and writers' work without a license takes value from the people who made it.

The four unions filed a joint amicus brief dated October 29, 2026 in In re: OpenAI, Inc., Copyright Infringement Litigation, No. 25-md-3143, before Judge Sidney H. Stein in the Southern District of New York. The document is filed as a proposed brief of amici curiae in support of the news organizations and authors already suing OpenAI, and runs to 5,998 words by its own certificate of compliance.

The unions are not asking for damages or an injunction of their own. They are asking the court to keep fair use a case-by-case question. "The harms of unlicensed training fall not just on companies like the News Plaintiffs, but on the countless professionals behind the work," the brief says. Its conclusion asks the court to "reject Defendants' fair use defenses."

The argument turns on what the model is built to do. SAG-AFTRA, AFM, WGAW and WGAE tell the court that the purpose of training on human work here is to generate work that emulates and competes with it, and they quote the Copyright Office's May 2025 training report, which rejected the claim that AI training "is inherently transformative because it is not for expressive purposes" as "mistaken." The brief leans on Andy Warhol Foundation v. Goldsmith for the proposition that the same copying can be fair for one purpose and not another, and on Thomson Reuters v. Ross Intelligence, the Third Circuit's September 29, 2026 ruling that training a competing legal research tool on Westlaw headnotes was not fair use.

The unions' distinctive contribution is the bargaining table. SAG-AFTRA's 2023 agreements set compensation terms around digital replicas, the 2023 WGA agreement established that AI is not a writer and cannot take credit or residuals from one, and AFM added AI safeguards for musicians in its 2024 agreement. The 2026 WGA and SAG-AFTRA negotiations produced notice-and-bargaining rights for third-party licenses of covered work used to train commercially available generative AI systems. Those terms only make sense, the brief argues, if training is a use that can be licensed. "A ruling here that training is categorically fair use would render moot the terms the Unions and the industry agreed to at the bargaining table," it says, and would harm "not just the copyright owners, but the people who create the work."

Those numbers are the unions' own filings: SAG-AFTRA says it represents more than 160,000 media professionals, AFM more than 80,000 musicians, and WGAW about 11,000 writers. OpenAI has not commented publicly on the brief, and no outlet reporting on the filing says it did.

The brief lands in a consolidated case, and an appellate court has already ruled on the fair use question once. The Third Circuit's Ross ruling affirmed that a licensing market for training data counts as a market even where none existed before the defendant's conduct, but the opinion expressly reserved the question of generative models. The OpenAI litigation is being briefed on the same factor now.

For creators outside the guilds, the practical stake is whether licensing becomes the default route rather than a negotiation. If the court accepts a categorical exemption, the terms SAG-AFTRA and the WGA won in 2026 would still bind the studios they bargain with, because a union contract is a private agreement, but they would not create a lever against AI developers who train on the same work without signing one.

Summary judgment reply briefs in the OpenAI MDL are due November 6, 2026.