On September 16, 2026, the U.S. Court of Appeals for the Ninth Circuit rejected an attempt to expand the scope of Section 1202 of the Digital Millennium Copyright Act (DMCA) so that it would become a new basis for copyright liability in cases where AI models produce code missing copyright management information. The decision concerns the Doe v. GitHub lawsuit brought by anonymous GitHub contributors against OpenAI and Microsoft.
The two companies used code published on GitHub as part of the training data for large language models, along with billions of other works. The plaintiffs said that new code produced by these models could resemble their code, but appears without the author’s name, a copyright notice, or other copyright management information.
What did the court decide?
Section 1202 prohibits intentionally removing copyright management information from a protected work. But the court distinguished between removing that information from a work that originally contained it and creating a new work that did not include the information in the first place. According to the Electronic Frontier Foundation (EFF), the absence of copyright information from a new output does not, by itself, establish that a party unlawfully removed it.
The ruling does not eliminate copyright protection for programmers. If an AI model is used to unlawfully reproduce protected code, rights holders still have the option to bring copyright-infringement claims. The contractual claims brought by the plaintiffs against the AI companies also remain pending, according to the published article.
Why does this matter?
The practical importance of the decision lies in its refusal to turn a limited statutory provision into a broad new right that could reach lawful uses unrelated to the actual removal of copyright information. The EFF believes that accepting the plaintiffs’ theory would have exposed activities such as preparing artists to remix older works, adapting educational materials for teachers, reverse engineering to understand code, and search engines that index the web to costly lawsuits.
This possibility particularly affects independent developers. Large companies can afford lengthy litigation, while an independent programmer might be forced to settle in the face of substantial statutory damages even when the underlying use is lawful. From this perspective, the decision limits a path that could have added new legal risks to software development and model research.
Limits of the decision and remaining questions
The EFF describes the ruling as narrow but important; it does not resolve all disputes involving training models on code or their outputs, nor does it decide the contractual claims remaining in the case. The available article also does not present the full text of the judicial opinion or details of those claims, so the decision cannot be considered a comprehensive resolution of the dispute over using code to train AI.
The broader reading offered by the EFF is that courts should apply the rights that Congress actually established when addressing new technologies, rather than create additional rights through an expansive interpretation. This is an editorial conclusion based on the organization’s position, while the confirmed legal effect is specific: the absence of copyright information from a new work is not, by itself, sufficient to prove that it was unlawfully removed.