Authors' Copyright Suit Over Databricks' DBRX Language Model Survives Dismissal Bid

Authors suing Databricks and MosaicML over alleged copyright infringement tied to the companies' large language models cleared a key pleading hurdle on April 21, when Judge Charles R. Breyer of the Northern District of California denied defendants' motion to dismiss and motion to strike allegations relating to the DBRX series of LLMs.

The case, In Re Mosaic LLM Litigation (No. 24-cv-01451-CRB), is a putative class action brought by authors who allege that Databricks and MosaicML copied their books in connection with developing and training AI models. The specific fight resolved by this order concerned the DBRX line of LLMs: plaintiffs alleged copying occurred during early development of DBRX, while defendants argued that because the copying was not tied to DBRX's final training dataset, the infringement allegations were too attenuated from the model as a product to state a claim.

Judge Breyer rejected that framing. The court concluded that the complaint sufficiently alleged copying linked to DBRX itself, pointing to allegations about Databricks' conduct during development and to statements by Databricks employees that plausibly implied copying connected to the model. The court noted that properly determining the degree of attenuation between the alleged copying and the DBRX product would require evidentiary considerations outside the pleadings, and that factual disputes of that kind are not appropriate to litigate on a motion to dismiss. The court was explicit that defendants may ultimately prevail on the attenuation question, but that the allegations were sufficient at the pleading stage.

The DBRX direct infringement claim had a prior life in this litigation. The court had initially dismissed it but allowed plaintiffs to move to amend once discovery surfaced facts to support the allegations. Plaintiffs filed a Second Amended Consolidated Complaint reasserting the claim, and defendants moved again to dismiss and to strike all DBRX-related allegations. The court concluded plaintiffs had cured the prior deficiencies by directly tying their infringed works to DBRX and by providing supporting context for the employee statements.

Because the court concluded the claim was adequately pleaded, it also denied the motion to strike, reasoning that the allegations implicate the merits of plaintiffs' claims.