A MESSAGE FROM OUR SPONSOR

Greatness isn't achieved in an instant. It's tested until there is no question — only performance.

In just the past few months, conventional wisdom has crept into the A.I. wars—namely, that the use of copyrighted material to train A.I. models will eventually be blessed as “fair use.” But perhaps that conclusion is a touch premature. Yes, the courts may well end up there. (I’ve been saying that for years.) But the evidence that they’re already on that glide path is perilously thin. It rests almost entirely on two summer rulings—Bartz v. Anthropic and Kadrey v. Meta, both out of San Francisco—while conveniently ignoring decisions that cut the other way. Chief among them is Thomson Reuters’s lawsuit against A.I. startup Ross Intelligence, which is now teed up to become the first appellate word on the issue after a U.S. District Court rejected Ross’s fair use defense and ruled that the company had infringed Thomson Reuters’s copyrights earlier this year.

Still, there’s no mistaking the recalibration among the plaintiffs’ bar. Litigators pressing copyright cases against A.I. giants increasingly seem to be shifting away from “training” as their strongest hand, and pivoting instead to copyright claims based on outputs. Consider last week’s hearing in the consolidated lawsuit that book authors brought against OpenAI and Microsoft in New York.

I went to that hearing expecting to witness a spirited debate over substantial similarity—whether, for example, a prompt asking ChatGPT to outline the next Game of Thrones novel, complete with George R.R. Martin’s characters and locales, crossed the copyright infringement line. OpenAI has moved to dismiss the plaintiffs’ case as insufficiently pleaded. But, regardless, the question of what exactly might constitute infringement in the world of output never quite took center stage at the hearing. Instead, Judge Sidney Stein found himself refereeing a more procedural fight: what it would mean, schedule-wise, if plaintiffs were allowed to retool their theories midstream.

Both sides have been racing to complete discovery before February, with a summary judgment briefing set for next August, class certification in September, and a trial likely sometime in 2027. But as time drags on, the plaintiffs keep testing new angles, sometimes inspired by legal developments elsewhere. The latest gambit: claiming that OpenAI downloaded a pirated dataset of books, even if it never actually used the works to train its models. It was a not-so-subtle nod to Anthropic’s $1.5 billion Bartz settlement, and OpenAI and Microsoft’s lawyers warned it would gum up discovery. “This is now a radically different case,” defense attorney Michelle Ybarra protested. “It will send us back to the drawing board.”

A MESSAGE FROM OUR SPONSOR

Greatness isn't achieved in an instant. It's tested until there is no question — only performance.

Judge Stein, sympathetic but wary, didn’t disagree. Still, he seemed reluctant to fence the case too narrowly, mindful that whatever happens here could set the tone for how courts handle A.I. for years to come. “Why shouldn’t we address this all now?” he asked.

Justice moves at its own glacial pace, as anyone who’s ever watched a major case against a tech company knows. One day, the entertainment industry is storming the barricades; the next, the world has moved on, and the legal detritus of a bygone era—think the criminal prosecution of notorious file-sharer Kim Dotcom—limps along in the background like some half-forgotten Cold War spy. Artificial intelligence is clearly here to stay. But it’s worth remembering that these A.I. battles are in their infancy, with legal theories emerging and fault lines taking shape that have yet to be tested.

One area likely to draw more fire is the messy question of who has legal standing to object to A.I. outputs—an issue inflamed, of course, by OpenAI’s unveiling of Sora 2, which sent Hollywood into a familiar panic, scrambling to figure out how (or even whether) to opt out. Here’s a scenario I can imagine landing in court: An actress decides she doesn’t want her likeness in the system. The studio that produced her movie, however, is perfectly comfortable letting A.I. users create fan works based on its intellectual property, perhaps even participating in OpenAI’s still-fuzzy plan to share revenue. The actress claims identity theft. OpenAI responds that it licensed the character fair and square. This is a collision that the entertainment industry has been sleepwalking toward.

Of course, the uneasy interplay between likeness and copyright isn’t new. In the 1990s, Paramount Pictures authorized a chain of airport bars modeled after Cheers—complete with animatronic versions of “Norm” and “Cliff.”  George Wendt and John Ratzenberger, who played the characters on the iconic NBC sitcom, promptly sued—arguing that the use of their doppelgängers amounted to a misappropriation of their likenesses. Paramount countered that the characters belonged to the studio, not the actors, and therefore weren’t protected by their rights of publicity. The Ninth Circuit Court of Appeals revived the case after a district judge’s initial dismissal, setting the stage for a jury to decide just how closely those robots resembled the actors—considering everything from costume to context. The case was settled before trial.

These days, of course, animatronics are just creepy antiques. Which studios will take the lead—and risk the legal exposure—in A.I. filmmaking? After all, David Ellison is looking to squeeze juice from his new acquisition and seems more than comfortable with the A.I. world. Notably, Paramount isn’t participating in Disney, Universal, and Warner Bros. Discovery’s copyright infringement cases against Midjourney, or their war against China’s MiniMax. If not Paramount, someone else will test the boundaries. Last year, Lionsgate licensed works—notably, franchise blockbusters John Wick and The Hunger Games—to the A.I. company Runway to use for training its models, much to the chagrin of the creative community.

Hollywood has been here before, if in less technologically sophisticated ways. When Crispin Glover declined to return as George McFly for Back to the Future Part II, the producers tried to mimic him by piling prosthetics on another actor. (Glover sued Universal and settled for a reported $760,000.) Less contentiously, when Paul Walker died in the midst of filming Furious 7, producers used C.G.I. to create a version of his face, which they imposed on body doubles played by two of Walker’s brothers. The Glover and Walker examples helped shape the collective bargaining talks between guilds and producers that eventually led to A.I.-era guardrails, such as the SAG-AFTRA rule that any A.I. simulation of an actor during production of a movie and television show triggers bargaining and payments.

But what happens when the simulation is sanctioned by the copyright holder, such as a studio, and spun up by a third party using Sam Altman’s latest algorithm? So far, I haven’t seen Hollywood talent contracts meaningfully updated. Many still contain catchall language granting rights “in any and all media now known or hereafter devised”—talk about a phrase begging for litigation. With another round of collective bargaining negotiations looming, maybe this will finally get its day at the table. More likely, the guilds and the producers will waltz around the A.I. licensing question and leave it to the courts, where, as ever, the wheels of justice turn however slowly.

Meanwhile, A.I. companies are stockpiling insurance coverage—hundreds of millions of dollars’ worth— to buffer against the coming wave of litigation, as reported by the Financial Times. You’ll be stunned to hear that the insurance industry is balking. And even if they were inclined to pay, it would never be enough. So companies like OpenAI and Anthropic are reportedly considering beefing up their litigation reserves, tapping investor funds to settle claims, and setting up their own “captive” insurance vehicles to manage the mess.

Even then, it’s hard to imagine what sort of Herculean feat would be required to tame the legal storm on the horizon. This isn’t just about the first- and second-wave lawsuits brought by creators—the “training” cases, the output cases, the in-the-style-of cases like Andersen v. Stability AI. It’s also about the nearly forgotten, but consequential, indemnification promises these companies made to their users. Microsoft president Brad Smith put it plainly: “If you are challenged on copyright grounds, we will assume responsibility for the potential legal risks involved.”

So far, with the notable exception of Elon Musk in that Blade Runner dustup, users themselves have largely avoided becoming targets. That won’t last. The deep-pocketed, heavily insured sports world, for instance, seems an obvious candidate to be dragged into court. In recent weeks, I’ve visited multiple NFL stadiums, and the folks running the Jumbotrons are already leaning hard on generative A.I. for hype videos and crowd-pleasing graphics. It doesn’t take much imagination to see where this is headed: a few errant pixels, a rights-holder with a good lawyer, and the opening act of what could be the longest and most expensive legal show on turf.

Thanks, Eriq. I’ll see everyone on Thursday.

Matt

Need help? Review our FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.

You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.

Puck is published by Heat Media LLC. 107 Greenwich St, New York, NY 10006