In Hazy World of AI Law, Judicial Skepticism of AI Output Infringement Claims Continues to Take Shape
In a concise order, District Judge Vince Chhabria of the U.S. District Court for the Northern District of California dismissed several claims set forth in comedian and author Sarah Silverman's class action copyright infringement lawsuit against Meta Platforms. Inc., captioned as Richard Kadrey, et al. v. Meta Platforms, Inc., No. 23-CV-03417-VC, 2023 WL 8039640 (N.D. Cal. Nov. 20, 2023) (“Silverman”). The case is the latest indication that courts are hesitant to accept broad claims of AI output infringement, particularly when allegations of substantial similarity are absent.
Sarah Silverman, Christopher Golden, and Richard Kadrey, on behalf of a proposed class of authors, sued Meta, claiming that Meta’s AI project, LLaMA, was trained on illegally acquired datasets that contained the authors’ works. Plaintiffs brought both direct and vicarious copyright infringement claims, alleging that because the LLaMA language models cannot function without information extracted from Plaintiffs' works, the language models are themselves infringing derivative works. Plaintiffs further argued that output of the LLaMA language models created in response to user prompts is based on information extracted from the Plaintiffs' works during the training process, and so, in Plaintiffs’ estimation, every output is therefore an infringing derivative work. In addition, the complaint alleged that any intermediate copying of Plaintiffs’ works that took place while training LLaMA constituted infringement. Plaintiffs also asserted a claim under the Digital Millennium Copyright Act (DMCA), arising from allegations that AI output omitted Copyright Management Information (CMI), as well as unfair competition, unjust enrichment, and negligence claims.
Meta moved to dismiss all claims except for the intermediate copying claim. The district court granted Meta’s motion in full and dismissed Plaintiffs’ claims, most with leave to amend.
Consistent with another recent Northern District ruling in Anderson v. Stability AI, No. 23-CV-00201-WHO, 2023 WL 7132064 (N.D. Cal. Oct. 30, 2023), which we reported on here, Judge Chhabria concluded that because the plaintiffs failed to allege any degree of similarity at all between their works and the AI output, the output could not constitute an infringing derivative work. Notably, the Court strongly disapproved of Plaintiffs’ direct infringement claim alleging that “LLaMA language models are themselves infringing derivative works.” The Court found this theory “nonsensical,” because “[t]here is no way to understand the LLaMA models themselves as a recasting or adaptation of any of the plaintiffs’ books.”
The Court dismissed Plaintiffs’ remaining “output” claims, finding that Plaintiffs had failed to allege facts setting forth a plausible violation of section 1202 of the DMCA, and that the unfair competition and unjust enrichment claims were preempted under section 301 of the Copyright Act. Plaintiffs were given leave to amend all dismissed claims, with the exception of a negligence claim that was dismissed with prejudice based on application of the “economic loss doctrine.”
In sum, Silverman represents yet another example of judicial skepticism regarding output infringement, and illustrates the difficulties that plaintiffs may have in successfully alleging such claims.