The United States Court of Appeals for the Ninth Circuit issued a published opinion in Doe v. GitHub, Inc., No. 24-7700, on 16 September 2026, affirming the dismissal of claims brought under the copyright management information provisions of the Digital Millennium Copyright Act. The matter reached the court as an interlocutory appeal certified by the United States District Court for the Northern District of California, which had dismissed those claims under Federal Rule of Civil Procedure 12(b)(6), first with leave to amend and then with prejudice. The named defendants include GitHub, Inc., Microsoft Corporation and a group of OpenAI entities.
Section 1202(b)(1) of Title 17 of the United States Code prohibits the intentional removal or alteration of copyright management information, and section 1202(b)(3) prohibits distributing copies of works knowing that the information has been removed or altered. The question certified under 28 U.S.C. section 1292(b) was whether those two subsections impose an identicality requirement. The panel held that the statute does not require literal identicality between the plaintiff's work and the allegedly infringing work, and treated identicality as a gloss on the statutory terms remove, alter and copies rather than as a separate element. Substantial reproduction without the information can serve as circumstantial evidence of removal, while material differences point instead to the creation of a new work. Applying that reading, the panel concluded that one who creates a new work and fails to include copyright management information has not removed or altered anything.
Providers of generative coding assistants, including GitHub, Microsoft and OpenAI, face a narrower route to liability under section 1202(b) where model output differs materially from the code the model was trained on. Authors who release code under open source licences requiring attribution notices cannot rest on section 1202(b) alone when the output they complain of is not a copy of their file. Claimants suing model providers in the Ninth Circuit now need to plead reproduction close enough to support an inference that the information was stripped from a copy, or to plead ordinary infringement, contract and licence breach instead.
The panel did not decide whether copyright management information is removed when licensed code is ingested as training data. The district court had recorded that the complaint was not about training and that the plaintiffs did not allege injury from the use of licensed code as training data, and the plaintiffs did not contest that characterisation, so the court treated the input theory as forfeited. The decision resolves the certified question only, and the remaining claims continue before the district court.
Licentium advises model providers, software businesses and rights holders on copyright and licensing exposure arising from machine learning systems. Work we undertake includes licence and attribution audits across open source dependencies, assessment of copyright management information risk in training and output pipelines, drafting of developer and customer terms for generative tools, and support in disputes over training data and model output.
Source: Doe v. GitHub, Inc., No. 24-7700 (9th Cir. 16 September 2026)