Writers Guild AI Deal Pushes Studios Down New Copyright Path

Movie and television studios will have to monitor their use of AI in the script-writing process or face copyright complications following the recent deal with the screenwriters’ union.

Provisions restricting—in some scenarios banning—use of AI in content fed to writers are embedded among other aspects of the collective bargaining agreement with Hollywood studios ratified by Writers Guild of America members Monday, less than two weeks after governing boards of the writers’ union ended the nearly five-month strike. The WGA also secured the right to bar the use of writers’ material to train AI models.

The AI requirements represent part of a broader push for studios to address an evolving area of copyright law with major implications for their ability to register copyrights and protect them.

The US Copyright Office and an appellate court have said that AI contributions to a creative work aren’t covered by copyright protections; only human-created work is, and any involvement by AI must be disclosed during the copyright registration process. Having a registration invalidated due to undisclosed AI involvement would, at least, force the company to drop the suit, register the work properly, and sue all over again—with limitations on remedies because the infringement predated the registration.

The guidance “puts the onus on studios to really be careful” about how they use AI, said Aaron J. Moss, a copyright and entertainment attorney at Greenberg Glusker Fields Claman & Machtinger LLP.

“You’re going to have these operating procedures that are going to be tweaked as a result of AI, to the extent that anybody is going to be using it,” Moss said. “Studios need to be disclaiming use of AI material that is more than de minimis. To the extent they don’t disclaim it, they run the risk later on if there’s a lawsuit.”

While the agreement at its core operates from a labor perspective, provisions restraining AI underscore creative groups’ belief that the technology often constitutes theft of not just jobs but creation, a bedrock copyright principle. The conditions under which AI infringes either by using creative works as training inputs or producing infringing outputs remain legally unclear and are the subject of ongoing litigation.

Tracking use of AI in a layered creative process like Hollywood screenwriting represents a new frontier given the Copyright Office’s stance on AI contributions. Errors in tracking AI through the script-writing and production processes could invalidate a copyright registration and derail a lawsuit. And maintaining such documentation may be difficult given ill-defined thresholds for what needs to be disclosed and how thoroughly.

Combating AI

The WGA ended its strike on Sept. 27 after the governing boards approved an agreement with major movie studio CEOs. The since-ratified deal would boost wages and implement numerous other benefits, including minimum staffing requirements and new streaming viewership-based bonuses.

It also includes restraints on AI use. The deal’s ban on deeming AI-generated “source material” means that it can’t be used to undermine a writer’s credit and associated rights. If a company provides any AI-generated material, it must disclose it to the writer.

Writers can’t be forced to use AI, according to the memorandum of agreement, though they can choose to if company policy allows. Companies, in turn, can reject AI—including where use could adversely affect copyrightability. The agreement also noted the uncertain and changing legal landscape but said nothing in the agreement restricts writers from asserting that their work can’t be used to train AI.

The guild’s goal was to protect members from AI interrupting their daily workflow or being used to replace them or drive down wages, said screenwriter John August, who was on the WGA’s negotiating committee.

“We want to make sure we’re not just protecting the notion of a writer, but what is human writing,” said August, who wrote the screenplays for the movies “Big Fish,” “Charlie’s Angels,” and the 2005 iteration of “Charlie and the Chocolate Factory.” “The origin of a thing does matter, and that’s what we’re codifying in there. Many of our members would consider that AI material stolen, and they don’t want to handle stolen goods.”

Copyright law looks at AI from a somewhat different angle than the labor agreement’s terms. The Copyright Office has wrestled with whether a machine can be a copyright holder—it recently determined that it can’t—and whether work created with the help of AI can be registered—it can but with caveats.

Creators have argued in lawsuits that AI infringes protected work when the work is incorporated into the mass quantity of material on which the model trains to enhance its capabilities. But courts are working through those issues in a barrage of lawsuits against OpenAI Inc., GitHub, and other AI companies.

Tracking Contributions

Errors could be costly. The Copyright Office ruled last month that failure to disclose and disclaim AI-generated aspects of a work would bar registration. In the context of a movie, that raises the question of how big an AI contribution would have to be to require a disclaimer—ranging from a piece of AI-generated artwork briefly in a background to entire chunks of dialogue.

There are existing processes to track contributions, including what is and isn’t protectable by the movie’s copyright—such as source material and historical facts—and those processes will now also have to track AI, said Moss, who he’s had clients use AI for set decorations. That makes tracking AI in the “chain of title”—including AI material given to or created by writers—critical, Moss said.

Studios should track any AI use that’s “more than de minimis” or risk a litigant invalidating the copyright registration—under a standard that is currently unclear, Moss said.

“It does remain to be seen how much detail is going to be given to the writers and how much detail is going to be kept ,” Moss said.

IP attorney Angela Kalsi of Greensfelder Hemker & Gale PC agreed that “studios would be foolish not to carefully track use of AI-generated materials.” Any legal dispute would be a “fact-intensive inquiry,” said Kalsi, who advises businesses on AI.

But the WGA agreement was designed as a bulwark against a flood of AI into the meatier parts of movie creation. Copyright and entertainment law professor Orly Ravid of Southwestern Law School said she sees good reasons for the WGA’s efforts to limit AI use.

“There are so many great applications of AI. I just don’t think there’s a need in human creativity,” Ravid said. “It’s not like it would make humankind better off. There’s already too much content.”

‘Going to Evolve’

The under-construction legal paradigms will ultimately interact with AI use under the new pact and beyond. The uncertainty makes it difficult to know how the WGA contract and studio use of AI will evolve, intellectual property attorney Aaron Swerdlow of Weinberg Gonser LLP said. AI technology is already “so far ahead of the law,” he said.

“We have no idea what the law’s going to say or what the technology is going to be able to do for the duration of the agreement,” Swerdlow said. “The technology, as it exists now, is going to evolve during the life of the WGA deal. We can’t address that now because no one knows what it’s going to look like.”

The Copyright Office laid down some markers in recent months, including by saying AI can’t be a creator—which courts have agreed with. But the office also said nothing bars AI from being “part of the creative process” given sufficient human creative input. Attorneys say that leaves a lot of gray area, especially with complex works like movies that have many creative hands touching myriad elements.

The prospect of AI-generated content being woven into various stages of the process further complicates an already convoluted area of law, Ravid said. Filtering out AI elements potentially adds yet another variable to often highly contested infringement and fair use questions.

“Copyright infringement is already messy, so I don’t think this is going to do it any favors,” Ravid said. “It’s such a complicated analysis. It’s going to be very difficult to distill out what is AI and what is not.”