Concord Music Group, Inc., et al. v. Anthropic PBC, et al. — Court Orders Meet and Confer on AI Output Copyright Claims After Hearing Reveals No Specific Infringement Allegations

Background

Concord Music Group and other music publishers sued Anthropic PBC and its CEO Dario Amodei, alleging copyright infringement related to the Claude AI model. The second amended complaint asserted, among other claims, direct copyright infringement with respect to outputs (Count II), alleging that Anthropic reproduced the publishers' lyrics in Claude's outputs, and it identified 20,517 works allegedly used to train Claude.

Anthropic moved to dismiss, arguing that the allegations were conclusory, failed to provide even a single example of an allegedly infringing output, and failed to identify which of the publishers' works were infringed in Claude outputs as opposed to Claude training. To avoid dismissal, the publishers argued that Anthropic copies their 20,517 lyrics in training and then outputs verbatim copies of those same lyrics, and that Anthropic generated AI output copying the lyrics to their 20,517 works verbatim and near-verbatim.

On November 4, 2026, the court heard argument on the motions to dismiss brought by Amodei and Anthropic. The court granted Amodei's motion but allowed the publishers to amend the complaint to cure the pleading deficiencies identified in that motion. With respect to Anthropic's motion, the court ordered the parties to meet and confer in light of the issues raised at the hearing, and the order provides additional guidance on the output-based claim.

The Court’s Holding

The court explained that in opposing Anthropic's motion, the publishers gave the clear impression that they had a factual basis for alleging that all 20,517 works allegedly used to train Claude were also reproduced in Claude's outputs, constituting separate and distinct acts of infringement, and the court assumed those assertions had some basis under Federal Rule of Civil Procedure 11(b)(3) and (4).

Counsel's statements at the hearing cast the complaint in a different light. It appeared the publishers asked the court to infer that all 20,517 works were infringed through Claude outputs merely by virtue of their inclusion in the training process. Yet the publishers had not alleged a single instance of any of the 20,517 works being reproduced as a Claude output, nor even that users prompted Claude for the lyrics to any of those works. Although the publishers invoked information and belief at the hearing, the complaint did not expressly make the output-based allegations on information and belief.

The court questioned whether information-and-belief pleading was permissible under these circumstances. In prior pleadings in the related case, the publishers had been able to identify specific acts of alleged output-based infringement, and the nature of the output-based claims implied that the publishers, like a third-party Claude user, could use Claude to generate infringing outputs, making it implausible that the information was solely in Anthropic's possession. The court concluded that the briefing did not accurately present the plausibility issue, and the parties shall continue to meet and confer on it.

Key Takeaways

• The court ordered the parties to continue to meet and confer on the output-based infringement issue and extended the deadline to file a status report to November 13, 2026, in light of the guidance and the Veterans Day holiday.
• Notwithstanding the ruling on Amodei's motion, the publishers need not file an amended complaint until Anthropic's motion to dismiss is resolved.
• The court's concern centered on the discrepancy between the briefing, which suggested a factual basis that all 20,517 works were reproduced in outputs, and hearing statements seeking inference of infringement merely from training-data inclusion.
• The publishers had not alleged a single specific instance of any work being reproduced as a Claude output, nor that users prompted Claude for the lyrics.
• The court doubted information-and-belief pleading was permissible where publishers had previously identified specific output-based infringements and could themselves use Claude to generate outputs.

Why It Matters

The order highlights that in AI copyright litigation, plaintiffs must plead a factual basis for output-based infringement rather than asking courts to infer it from training-data inclusion, and that courts will scrutinize whether briefing accurately frames the plausibility question under Rule 12(b)(6), particularly where, as here, the claim concerns generative AI outputs.

It also signals judicial skepticism toward broad information-and-belief pleading for AI output claims where the plaintiff may be able to investigate the system itself, which may require copyright plaintiffs to conduct more thorough pre-filing investigation of AI outputs.