Yesterday, Judge Eumi Lee held a marathon hearing to hear arguments regarding the plaintiffs’ motion for a preliminary injunction in Concord Music v. Anthropic, a copyright suit filed by music publishers against Anthropic for alleged infringement in the training and alleged regurgitated lyric outputs of Anthropic’s AI model, Claude. And by marathon I mean it ran for nearly 4 hours, leaving no time for argument on Anthropic’s motion to dismiss, which had to be rescheduled for another day.
How many lawyers does it take to litigate AI disputes? Each side had about 10 lawyers at the hearing; 5 lawyers sat at the counsel’s table for each side. Plus, it looked like a team of 5 lawyers unrelated to the case were also in the audience. So, the correct answer is at least 20.
Highlights: I will leave it to Bloomberg Law and others to report a full blow-by-blow account. Here are my general impressions in the order of how the issues were discussed. *I haven’t read the parties’ briefing on this motion, so my impressions were formed based on the hearing arguments.
1. Judge Lee devoted the first 90 minutes to argument on the scope of the preliminary injunction sought. She was clearly concerned about scope and said she is hoping the parties can hammer out the language of a stipulation to keep Anthropic’s current guardrails in place (so she wouldn’t have to issue an injunction). In fact, the beginning of the argument involved both sides relating how close to a stipulation they were: the music publisher plaintiffs’ are fine with Anthropic’s current guardrails (to stop lyrics being regurgitated on Claude), but didn’t like the proposed stipulation language to allow Anthropic to “modify” the guardrails for updated versions of Claude in a way that would lead to “degradation”; Anthropic said the proposed stipulation didn’t use the word “degradation” and they would of course implement future guardrails to continue to stop regurgitated lyrics. Judge Lee proposed some language that might stipulate the “effectiveness” of guardrails for new versions, while deleting the word “modify.” But she suggested they might discuss it after the hearing, since they had expended so much time on this issue.
2. In the grand scheme of things, the music publishers seem to be fighting mainly over (1) improved versions of Claude and (2) new untrained models in the future. The Plaintiffs are fine with the current guardrails for the current Anthropic. The Plaintiffs want guardrails on Anthropic’s updated versions of Claude (see dispute in 1 over effectiveness). They also want an injunction against future use of their works in future untrained models of Anthropic. If I heard correctly, both Anthropic and Judge Lee seemed to indicate (suggest) that future untrained models does not fall within the Plaintiffs’ motion. At the very least, Anthropic argued there’s no imminent injury to justify a preliminary injunction.
3. Judge Lee was concerned whether what the Plaintiffs’ sought was a mandatory injunction (instead of merely a prohibitory one freezing the status quo); a mandatory injunction generally raises far greater concerns for courts. She was also concerned about whether the plaintiffs had shown an irreparable injury to justify an injunction, or whether the injuries alleged were remediable by damages at law (e.g., lost licensing fees or statutory damages). Other issues related to scope that concerned the Judge: (1) how many musical works would be subject to the injunction–hundreds of thousands, if not millions? (2) how would enforcement of the injunction be implemented in terms of the effectiveness of guardrails and would it also extend to API versions of Claude with 3rd parties?
4. Judge Lee indicated she thought the fair use in the training of AI models is a novel, cutting-edge question that pushes the boundaries of fair use and copyright law. (Those words were her words if I remember correctly, although I am combining different instances during the hearing in which she expressed her view on the legal issue.) However, both sides struck me as arguing that fair use was more cut-and-dried based on the prior cases they respectively cited. At the hearing, Anthropic cited the Google Book Search 2d Circuit decision, the 9th Circuit Sega decision, and the 4th Circuit iParadigms decision. The music publishers cited the 9th Circuit Napster decision and the Supreme Court’s Andy Warhol Foundation v. Goldsmith decision, and distinguished Sega.
If I heard correctly, Judge Lee wasn’t convinced that the Andy Warhol precedent disposed of the fair use defense in this context, at least not as strongly as the music publishers suggested. This is consistent with the novel issue (noted above).
5. I was quite surprised that Google v. Oracle–a case involving “a new and transformative [computing] program,” 593 U.S. 1, 40 (2021)–didn’t receive much, if any, discussion at the hearing, particularly as to its relationship with Andy Warhol . We do have one district court decision attempting to interpret both cases in the context of AI training, but no one discussed it. Judge Stephanos Bibas concluded one shouldn’t “overread” Andy Warhol, and that Google has relevance as well given its greater similarity as a technological case. See Thomson Reuters Enter. Centre GMBH v. ROSS Intelligence Inc., 694 F. Supp. 3d 467, 482 (D. Del. 2023). While Judge Bibas is now considering renewed motions for summary judgment, his opinion is still a published precedent.
Judge Bibas explained:
“Ross’s uses were undoubtedly commercial. And one of its goals was to compete with Westlaw. Thomson Reuters contends that this commercial use weighs heavily against finding fair use. In support of this, it cites the Supreme Court’s recent decision in Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 143 S. Ct. 1258 (2023). There, the Court determined that the use in question was not fair largely by emphasizing its commercial nature. See id. at 1279–80. But I decline to overread one decision, especially because the Court recognized that “use’s transformativeness may outweigh its commercial character” and that in Warhol, “both elements point[ed] in the same direction.” Id. at 1280. Plus, just two terms ago, in a technological context much more like this one, the Court placed much more weight on transformation than commercialism. Google, 141 S. Ct. at 1204 (“[A] finding that copying was not commercial in nature tips the scales in favor of fair use. But the inverse is not necessarily true, as many common fair uses are indisputably commercial.”). So I focus on transformativeness.”
Putting aside the substance of Judge Bibas’s interpretation, I think he’s definitely correct in wrestling with the relationship of the two most recent Supreme Court decisions on fair use, Google and Andy Warhol. If, as likely, the Supreme Court will consider whether AI training involves a fair use in one or more of the copyright lawsuits now pending, the Supreme Court will do the same.