Concord Music Group, Inc. v. Anthropic PBC

District court grants motion to dismiss music publishers’ claims of contributory and vicarious copyright infringement with leave to amend, and denies publishers’ motion for preliminary injunction, finding that allegations that unidentified users might prompt AI product to produce copyrighted lyrics were insufficient to establish third-party infringement and that injunction sought was overbroad and not necessary to prevent irreparable harm. 

Plaintiffs are a group of music publishers, led by Concord Music Group Inc. and Universal Music Corp., that brought claims for direct and secondary copyright infringement and removal of copyright management information (CMI) against Anthropic PBC stemming from the use of Anthropic’s signature AI product line, Claude. Plaintiffs alleged that Anthropic infringes their exclusive rights to musical lyrics by using their works to train Claude, and by distributing the lyrics in Claude’s outputs when Claude is prompted. Plaintiffs sought a preliminary injunction to prohibit Anthropic from using the copyrighted lyrics for training purposes. Anthropic moved to dismiss plaintiffs’ claims for contributory and vicarious infringement and the intentional removal of CMI, and opposed plaintiffs’ motion for a preliminary injunction. The court found in favor of Anthropic in two consecutive decisions: first, denying the motion for a preliminary injunction as overbroad and failing to establish irreparable harm, and second, granting Anthropic’s motion to dismiss with leave to file an amended complaint. In granting the motion to dismiss, the court found that plaintiffs failed to show any direct third-party infringement, i.e., that users submitting prompts actually generated copies of copyrighted lyrics, and that Anthropic knew or had reason to know of any third-party infringement. The court also found that plaintiffs did not sufficiently allege that Anthropic acted intentionally as to the removal of CMI, particularly given plaintiffs’ acknowledgment that CMI was not removed across the board throughout Claude’s training process. 

Prior to the court’s decisions, the parties entered into a joint stipulation by which Anthropic agreed to maintain already implemented guardrails as to its current AI models and products. The court noted that the stipulation resolved plaintiffs’ motion for injunctive relief to the extent that it concerned Claude’s output. Accordingly, the court’s denial of the motion for a preliminary injunction focused solely on the alleged input-based infringement related to using copyrighted lyrics in training Claude. The court first found that the requested injunction was overbroad and poorly defined, as it would ostensibly cover hundreds of thousands of works, including future acquisitions, with no clear way for Anthropic to ascertain or comply with its terms. The court expressed concerns about enforceability and undue burden on Anthropic. The court next found that plaintiffs failed to establish irreparable harm in the form of reputational or market-related injury. Because the evidence of reputational harm was primarily derived from Claude’s outputs, the court found that such harm was already addressed by the parties’ stipulation. Further, plaintiffs provided no evidence of lost deals or unfavorable terms in the existing licensing markets, and plaintiffs’ claims regarding the impact on the emerging AI training market were speculative. In light of plaintiffs’ failure to establish the prerequisite of irreparable harm, the court declined to address the remaining factors in the preliminary injunction analysis or the merits of the underlying infringement claims.

One day after its denial of the motion for injunctive relief, the court granted Anthropic’s motion to dismiss plaintiffs’ claims for contributory and vicarious infringement and the intentional removal of CMI. With respect to the contributory infringement claim, the court held that the allegations that unidentified users “might” use Claude to seek the copyrighted lyrics were insufficient to state a claim. Moreover, at hearing on the motion, plaintiffs appeared to acknowledge that at least some of their examples of prompts that yielded copyrighted lyrics were performed by plaintiffs or their investigators rather than by actual users. The court did not opine on whether such prompts by plaintiffs’ investigators would constitute a predicate act of third-party infringement, primarily because the complaint made only a generic reference to “users” whose prompts resulted in infringing outputs.  

With respect to the vicarious infringement claim, the court found that plaintiffs adequately alleged that Anthropic received a direct financial benefit from the allegedly infringing activity, based on the allegation that Anthropic is paid each time Claude is prompted and each time Claude generates an output. Nevertheless, the court dismissed the vicarious infringement claim due to plaintiffs’ failure to allege direct infringement by a third party. Finally, the court found that plaintiffs failed to allege that Anthropic intentionally removed CMI or knew that such removal would induce infringement. In addition to finding the allegations on this point conclusory, the court found the allegations were contradicted by plaintiffs’ acknowledgment that some of Claude’s outputs include attribution for lyrics. Such an allegation indicates that CMI is not removed from each of the works during Claude’s training process, and undermines plaintiffs’ argument that Anthropic intentionally removed CMI.

The court allowed plaintiffs 30 days to file an amended complaint, but prohibited plaintiffs from adding any new claims without leave of court.

Summary prepared by Safia Gray Hussain and Chloe Gordils