It’s been a busy two weeks for copyright and AI cases.Thaler v. Perlmutter, No. 23-5233 (D.C. Cir. March 18, 2025)
This case involved the completely AI-generated work, “A Recent Entrance to Paradise”:
The copyright applicant, Thaler, disclaimed any human involvement in the work’s creation. The district court ruled it wasn’t copyrightable.
The DC Circuit affirms, saying that “the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being.” By anchoring its ruling in the statute, the court sidesteps whether this outcome is Constitutionally required. The court also sidesteps the more interesting question: when a human interacts with AI to produce a work, is the work entirely protectable, protectable only as to the human-assignable contributions, or unprotectable? The court says talismanically that “the human authorship requirement does not prohibit copyrighting work that was made by or with the assistance of artificial intelligence,” but leaves open how that will translate in practice. (I’m not sure what “by” modifies in that sentence).
Bonus: The opinion has a shoutout to Data’s wonderful Ode to Spot from Star Trek: The Next Generation. It’s an apropos reference because Data is an android who sometimes creates putatively copyrightable works, and one of the best-loved series episodes turns on whether Data should be given the same legal rights as humans. (An even more on-point reference would have been the overlooked Star Trek: Voyager episode Author, Author, when The Doctor, a computer-generated hologram, claims copyright in his work). The court says:
Star Trek’s Data might be worse than ChatGPT at writing poetry, but Data’s intelligence is comparable to that of a human being. See Star Trek: The Next Generation: Schism (Paramount television broadcast Oct. 19, 1992) (“Felis catus is your taxonomic nomenclature, an endothermic quadruped, carnivorous by nature”).
I did a quick search and this appears to be the first time the term “endothermic quadruped” has appeared in the Westlaw and Lexis caselaw database.
For Chanukah 2023, I gave my kids an ornament with Data reciting part of the ode! It was a big hit.
Concord Music Group, Inc. v. Anthropic PBC
1) Opinion denying a preliminary injunction, 2025 WL 904333 (N.D. Cal. March 25, 2025)
Anthropic previously agreed to maintain its guardrails designed to prevent “output that reproduces, distributes, or displays, in whole or in part, the lyrics to compositions owned or controlled by Publishers, or creates derivative works based on those compositions.” (The “in part” language is funky, because surely a line or two of lyrics constitutes fair use).
Having done so, the only remaining equitable issue is the use of the copyrighted works for training purposes. The plaintiffs wanted to block all of their copyrighted works from being indexed, but this is an ever-changing corpus of hundreds of thousands of works. The court responds that “there is no manageable process through which the Court could enjoin the use of the Works for training purposes without unduly burdening Anthropic.”
The court also isn’t sold on the irreparable injury. First, “Publishers have not demonstrated reputational harm based on the use of the Works as training input.” The plaintiffs cited their loss of control, but the court responds: “To a certain extent, all copyright infringement involves a loss of control. But accepting Publishers’ argument on this record would imply automatic entitlement to injunctive relief in all copyright cases, a premise the Ninth Circuit has rejected.”
Second, the court doesn’t accept that indexing irreparably harms the market for the works:
Publishers have not submitted any evidence that Anthropic’s use of lyrics to train Claude reduces license fees with lyric aggregators, lyric websites, or other existing licensees, which provide entirely different services and do not compete with Claude. To the extent Publishers are arguing that lyrics appearing in Claude’s output undermines their ability to license to lyrics aggregators, this potential harm is addressed by the parties’ stipulation. In addition, Anthropic vigorously disputes that Claude could be used as an alternative to lyrics aggregators, as Claude was designed precisely not to perform these functions…
Publishers have not identified any lost licensing deal or any licensing arrangement that Publishers have had to renegotiate on less favorable terms.
The court also notes that money damages could compensate for any market harms.
In a footnote, the court distinguished the Thomson Reuters v. ROSS opinion because “(1) it addressed the merits of the parties’ respective infringement claims and defenses at the summary judgment stage; (2) it did not concern a generative AI model; and (3) the parties in that case were direct competitors.”
The court concludes: “By seeking a preliminary injunction, Publishers are essentially asking the Court to define the contours of a licensing market for AI training where the threshold question of fair use remains unsettled.” The court wasn’t willing to do this.
2) Opinion granting Anthropic’s motion to dismiss some claims, 5:24-cv-03811-EKL (N.D. Cal. March 26, 2025)
The next day, the court dismissed the contributory copyright, vicarious copyright, and CMI removal claims against Anthropic, with leave to amend.
Contributory Copyright Infringement. “Publishers have not clearly alleged a predicate act of direct third-party infringement.” Also, “the inclusion of protected lyrics in Claude’s training corpus alone does not establish actual knowledge by Anthropic of third-party infringement.”
Vicarious Copyright Infringement. Same problem with the lack of a predicate third-party infringer. However, the court says there’s sufficient allegations of direct financial benefit because the copyrighted works in the index act as a draw for paying customers.
CMI Removal. The plaintiffs “have not sufficiently pled that Anthropic acted intentionally as to the removal of CMI.” The court distinguishes the Github ruling because, in that case, “the court concluded it could reasonably infer intentional removal of CMI based on the allegations that (1) the defendants knew the source code they used to train their AI programs contained CMI, (2) the defendants knew that CMI was important for protecting copyright interests, and (3) Github had regularly processed takedown notices “such that it was aware its platform was used to distribute code with removed or altered CMI in a manner which induced infringement.””
New York Times Co. v. Microsoft Corp., 23-cv-11195 (SHS) (OTW) (S.D.N.Y. March 26, 2025)
The court smmarized its rulings on multiple topics, but says a written opinion detailing its reasons is forthcoming:
the Court hereby denies (1) OpenAI’s motions to dismiss the direct infringement claims involving conduct occurring more than three years before the complaints were filed; (2) the defendants’ motions to dismiss the contributory copyright infringement claims; and (3) the defendants’ motions to dismiss the state and federal trademark dilution claims in the Daily News action.
The Court grants defendants’ motions to dismiss the common law unfair competition by misappropriation claims and OpenAI’ s motion to dismiss the “abridgment” claims in the CIR action, and dismisses each of those claims with prejudice.
With respect to the DMCA claims, the Court grants (1) Microsoft’s motions to dismiss the section 1202(b )(1) claims against it in all three actions, (2) OpenAI’ s motion to dismiss the section 1202(b)(l) claim against it in The New York Times Company v. Microsoft Corporation, et al., No. 23-cv-11195 (“the Times action”), and (3) defendants’ motions to dismiss the 1202(b )(3) claims against them in all three actions, and dismisses each claim without prejudice. The Court denies OpenAI’ s motions to dismiss the 1202(b)(1) claims against it in the Daily News and CIR actions
The explanatory opinion should be a doozy.
