OpenAI Bid to Centralize IP Suits Follows Discovery Setbacks

OpenAI Inc.'s effort to centralize eight copyright suits into multidistrict litigation could be a tactic to mitigate a recent ruling that it must face a key claim brought by Intercept Media Inc. and slow down some of the more advanced proceedings, attorneys say.

Judge Jed S. Rakoff on Nov. 21 refused to dismiss Intercept Media’s claim that OpenAI intentionally removed copyright management information from the news outlet’s work in violation of the Digital Millennium Copyright Act to use it to train ChatGPT. His order in the US District Court for the Southern District of New York diverges from other decisions by judges in the same district and California that have dismissed similar claims against the artificial intelligence company.

The company is embroiled in discovery fights over its executives’ social media messages, and it’s sought, unsuccessfully, to obtain revenue data, and reporters’ notes from some plaintiffs in the cases.

OpenAI informed S.D.N.Y. Magistrate Judge Ona T. Wang of its decision to petition the Judicial Panel on Multidistrict Litigation at a Dec. 3 status conference and two days later filed a letter in the US District Court for the Northern District of California to inform it of its decision.

MDL centralizes pretrial activities such as discovery in cases that involve common factual questions. Though it could be more procedurally efficient for an MDL court to handle all the pretrial proceedings for the eight lawsuits—including those brought by New York Times Co. and authors like comedian Sarah Silverman—the MDL court could also potentially decide dispositive motions like summary judgment, Santa Clara University School of Law professor Edward Lee said. That “would be quite a change from the multiple judges” in New York and California to just one judge pre-trial, he said.

But consolidating discovery across the cases “would ultimately slow down litigation since the cases are at different points and some would have to catch up,” said Mark Lemley, an intellectual property professor at Stanford Law School who also represents Meta Platforms Inc. and Stability AI in copyright lawsuits over training data. Some plaintiffs have opposed OpenAI’s previous motions to merge certain lawsuits and would have an opportunity to argue against an MDL petition as well.

OpenAI’s bid for MDL “could be an end run around Judge Rakoff’s opinion that is out of step with all these other decisions,” Lemley said. OpenAI can say to the MDL panel, “Now we’re facing inconsistencies in decisions and we want these consolidated,” he said.

The New York Times declined to comment, noting OpenAI’s petition hasn’t been filed.

“We don’t have a comment on this one outside of what’s in the letter itself,” OpenAI said in an emailed statement.

As time has gone on, judges that receive MDL jurisdiction have gained more power to resolve motions to dismiss and motions for summary judgment, said Michael Kaufman, dean and professor at Santa Clara Law. The judge could go as far as holding a bellwether trial, he added, as a test case for the most contested issues.

Resting the power in one judge’s hands could potentially disadvantage the plaintiffs in other ways too, James Gatto from Sheppard Mullin said. For example, it could limit the number of “bites at the apple” they get by testing different theories in different courts.

“Many times when you’re trying to address novel legal issues like this, you often see plaintiffs take a variety of tactics, raise different issues in different cases, to kind of see what sticks,” he said.

MDL Assignment

Because assigning the case falls to the MDL panel—not the litigants—the parties risk uncertainty regarding the judge or judges who will be conducting these proceedings, Kaufman said. There’s no guarantee OpenAI will land a favorable judge in MDL.

The judicial panel could even designate Rakoff, Lemley added.

OpenAI may also be seeking an MDL to limit each of its witnesses to one deposition instead of having them face separate depositions for each lawsuit, Santa Clara’s Lee said.

To successfully centralize all the cases in one court, OpenAI will have to show there are common questions of fact across the cases, and that the transfer would be convenient and cost-effective. Although each of the cases center on different types of content—news, books, images—it could be argued they all involve protected material that OpenAI is alleged to have copied to train AI systems, Gatto said.

“The facts don’t have to be identical for an MDL,” he said, “but they need to be similar.”

Centralizing the cases could create complications, though, especially with the novel questions at issue. If the cases return to their original courts for trial, it would open the possibility of creating conflicting precedent. Absent an interlocutory appeal while the case was still consolidated, the parties would raise any objections in a post-trial motion in the separate courts, Fabio Marino of Womble Dickinson Bond said. Appeals of any post-trial rulings could lead to the same question being appealed to different circuit courts.

“Theoretically there could be different decisions from different appellate courts on basically the same issue,” Marino said.