Judge Breyer just granted Mosaic ML’s partial motion to dismiss the newly added copryight claim in the First Amended Complaint against Databricks based on its DBRX large language models. And the Judge did so without further leave to amend.
Judge Breyer held that the plaintiffs’ general allegation that, upon information and belief, Databricks used pirated books from shadow libraries did not state a claim based on factual allegations.
This is a major win for the defendant AI companies.
Plaintiffs allege on information and belief that Databricks trained its DBRX models on open data sets that included shadow library websites with Plaintiffs’ works, FAC ¶ 41, but even allegations on information and belief must “be supported by a ‘factual foundation leading to that belief.’” Reaper v. Ace Am. Ins. Co., No. 21-cv-5876-HSG, 2023 WL 1478474, at *2 (N.D. Cal. Feb. 2, 2023). It is the lack of factual, rather than conclusory, allegations as to what materials Databricks actually used to train the DBRX models that dooms Plaintiffs’ direct infringement claim against Databricks.
–Judge Breyer
Here are the allegations in the FAC that Judge Breyer found insufficient to state a claim:
