Ninth Circuit rejects the DMCA theory in Doe v. GitHub

On September 16, 2026, the Ninth Circuit affirmed the dismissal of the DMCA Section 1202 claims against GitHub, Microsoft, and OpenAI over the Copilot and Codex coding tools (Doe v. GitHub Inc., No. 24-7700, affirming Judge Jon Tigar’s dismissal below). The court held that the tools do not remove copyright management information from existing works. They generate new works that never contained that information in the first place. The panel reasoned that treating substantially similar output as a DMCA violation would supplant traditional copyright law and expose defendants to the DMCA’s enhanced statutory damages. The court expressly declined to decide whether similarity between AI-generated code and copyrighted code supports an ordinary infringement claim, so that question remains open. The breach-of-contract claims over the open-source licenses remain pending in the district court. Reuters coverage

The practical point: the enhanced-damages shortcut is gone. Code that is substantially similar to licensed open source is still analyzed as plain infringement, and whether open-source licenses are enforceable as contracts against AI training and output is now the live core of the case.

Three more points from the opinion are worth knowing (Doe v. GitHub, Inc., No. 24-7700, slip op. at 13-17 (9th Cir. Sept. 16, 2026)):

• No identicality requirement. The court rejected the district court’s rule that a Section 1202 claim requires an identical copy. Minor cosmetic changes do not protect a defendant who substantially reproduces a work and strips its copyright management information. The line the court drew is generation versus copying, not identical versus modified.
• Training was not decided. The panel decided only the output theory. Whether removing copyright management information from code before using it as training data violates the DMCA was not preserved below and remains an open question.
• Architecture matters. The court noted that a retrieval-based tool that surfaces stored material could present a stronger basis for a removal claim than a generative one. A tool’s architecture now bears on its DMCA exposure.

This page reflects current U.S. copyright law and general international principles. It is provided for educational purposes and does not constitute legal advice.