IP Law Daily, COPYRIGHT—N.D. Cal.: Authors’ infringement claims against Databricks over DBRX series of AI models survive motion to dismiss, (Apr 23, 2026)

By Steven Melendez

A court ruled that allegations of copying are sufficiently linked to the AI model.

A federal judge declined to dismiss a direct copyright infringement claim brought by authors against tech companies Databricks and MosaicML for copyright infringement related to a DBRX series of AI models, finding the authors sufficiently alleged infringement related to the models.

The authors have alleged that the tech companies copied books from so-called “shadow libraries,” which are online repositories allegedly full of pirated works, in order to develop the large-language models (LLMs) used for generative AI. Initially, the court had dismissed an added infringement claim related to a series of LLMs known as DBRX, but noted the authors could move to amend their complaint if and when discovery revealed a basis for factual allegations for the claim (In Re Mosaic LLM Litigation, No. 3:24-cv-01451-CRB (N.D. Cal. Apr. 21, 2026)).

The authors indeed later filed a second amended consolidated complaint reasserting a DBRX-related direct infringement claim, with the court at the time noting that the added allegations shouldn’t be a surprise to the tech companies, since the authors had already sought to include variations on them. Databricks and MosaicML, who had opposed allowing the added claim, soon moved to dismiss.

Motion to dismiss. Databricks and MosaicML argued that the authors’ allegations about infringement are not specifically linked to DBRX. “Essentially, because Plaintiffs only allege that Databricks copied their books during early development of DBRX and not its actual training, Plaintiffs fail to state a claim as to DBRX,” the defendants contend.

But the authors argued that their allegations are still directly tied to the DBRX models “and that copying is copying—no matter at what stage in LLM training,” the plaintiffs contend.

A direct copyright infringement claim, according to the court, requires a pleading that “at least one exclusive right granted to copyright holders” is violated. Here, the court found, the complaint sufficiently made such an allegation. The exact details of the claim are redacted in the unsealed version of the court order, though the order alludes to employee statements. The tech companies argued those statements “were previously deemed insufficient to state a claim and should again fail,” but the authors “directly tie their infringed works to DBRX, and the employee statements provide supporting inferences when read in context, particularly when viewed alongside other more direct statements.”

The tech companies also argued that because the authors don’t allege copying in “DBRX’s final training dataset,” the infringement allegations are too distanced from the DBRX product, arguing alleged copying in development doesn’t support a claim for direct infringement in the AI model itself. But, the court found, the complaint does sufficiently tie the copying to DBRX.

“But the complaint, as discussed above, does tie the copying to DBRX itself,” the court found. “Properly determining the degree of attenuation would require evidentiary considerations outside of the pleadings.”

The court found the factual disputes inappropriate to decide at the motion to dismiss stage, allowing the claim to proceed.

The Case is No. 3:24-cv-01451-CRB.

Judge: Breyer, C.

Attorneys: Cadio R. Zirpoli (Saveri Law Firm, LLP) for Stewart Onan. Jedediah Wakefield (Fenwick & West LLP) for Databricks, Inc. and Mosaic ML, Inc.

Companies: Databricks, Inc.; Mosaic ML, Inc.

MainStory: TopStory Copyright CaliforniaNews TechnologyInternet GCNNews