Recommended Articles
Legal & Policy

Doe v. GitHub: The Ninth Circuit Rules AI-Generated Code Is Not a DMCA Violation Just Because It Lacks Attribution

By Angeled · October 10, 2026 · Legal & Policy
Doe v. GitHub: The Ninth Circuit Rules AI-Generated Code Is Not a DMCA Violation Just Because It Lacks Attribution
Doe v. GitHub: The Ninth Circuit Rules AI-Generated Code Is Not a DMCA Violation Just Because It Lacks Attribution

On September 16, 2026, the U.S. Court of Appeals for the Ninth Circuit delivered one of the most consequential AI copyright decisions to date. In Doe v. GitHub, Inc. (No. 24-7700), the court affirmed the dismissal of Digital Millennium Copyright Act claims brought by anonymous programmers against GitHub, Microsoft, and OpenAI — holding that an AI coding tool generating new code without copyright management information does not "remove or alter" that information from an existing work, as DMCA Section 1202(b) requires. Legal analysis of the ruling published this week (October 9) unpacks why the decision matters for every developer building on AI.

What the case was about

A group of programmers who published open-source code in public GitHub repositories sued over GitHub Copilot and OpenAI's Codex, AI tools trained on publicly available code. They alleged the tools sometimes generated verbatim or near-verbatim portions of their work without the attribution, copyright notices, or license terms that accompanied the originals — and framed the claim under the DMCA's copyright management information (CMI) provisions rather than as ordinary copyright infringement. The framing was strategic: the DMCA allows statutory damages of up to $25,000 per violation, and it converts every AI-generated snippet into a potential standalone offense.

The plaintiffs advanced two theories. The "input" theory claimed GitHub and OpenAI stripped CMI from licensed code before feeding it into training data. The "output" theory claimed Copilot's generations, which reproduce their code without attached CMI, constitute removal of that information.

What the court held

The Ninth Circuit rejected the output theory on a deceptively simple ground: Copilot and Codex do not remove CMI from a copy of an existing protected work. Instead, the court found, they generate new works that never contained the information in the first place. In the panel's view, the plaintiffs were asking the court to "transform run-of-the-mill copyright infringement claims into DMCA claims" — an invitation it declined.

Two nuances in the decision matter enormously:

  • The court rejected an identicality requirement. The district court had dismissed the claims because plaintiffs could not show outputs were "identical" to their code. The Ninth Circuit took a different path, declining to impose that strict standard — meaning a future case with different facts could come out differently.
  • The input theory was forfeited. The panel refused to even consider the training-time stripping claim because the plaintiffs had failed to correct the district court's repeated statements that the case was not about training data. The question of whether removing CMI during model training violates Section 1202(b) remains entirely open.

Judge Eric Miller, writing for the panel, also made clear the decision was not a clean bill of health for AI coding tools. The court expressly declined to address whether the substantial similarity between Copilot outputs and existing code could support ordinary copyright infringement claims — and the district court allowed the plaintiffs' breach-of-contract claims (rooted in open-source license terms) to proceed.

Background: why Section 1202(b) became the battlefield

Section 1202(b) of the DMCA, enacted in 1998, prohibits the intentional removal or alteration of copyright management information — things like the title, author identity, copyright owner, and terms of use — when done knowing it will facilitate infringement. It was designed for the era of copied files with stripped headers, not for generative models that synthesize new text from statistical patterns.

That mismatch is exactly what the Ninth Circuit addressed. Generative AI does not "copy" in the traditional sense: it produces novel sequences that may resemble training data without containing any lineage information at all. Treating every such output as a CMI "removal" would have created strict-liability exposure for virtually every deployed AI model — a flood of litigation the court was unwilling to open.

Why it matters for founders and investors

For AI startups, the ruling removes a large contingent liability from the balance sheet — at least in the Ninth Circuit, which covers much of the tech industry. But three risks remain live:

  1. Ordinary copyright infringement is untouched. Substantial-similarity claims and license-based contract claims still proceed. Founders building code assistants should invest in output-filtering and attribution features, not just legal arguments.
  2. Training-time CMI stripping is unresolved. The input theory died on a procedural forfeiture, not on the merits. Data-pipeline hygiene — preserving license metadata through training-data preparation — remains legally prudent.
  3. The breach-of-contract path is open. Open-source license terms (like MIT and GPL attribution requirements) are contractual obligations independent of the DMCA. A startup can win the DMCA fight and still lose on license compliance.

What to watch

Expect further appellate litigation testing the input theory with better-preserved claims, and watch whether other circuits follow the Ninth Circuit's "new works" reasoning. For founders, the practical move is to treat this as a ceiling on one theory of liability, not a floor of safety: build attribution and provenance tooling into AI coding products now, before the next case forces it.

Sources: Sheppard Mullin analysis (https://www.sheppard.com/insights/blogs/the-ninth-circuit-affirms-dismissal-of-dmca-cmi-removal-claim-against-ai-code-generator-output-is-a-new-work-not-a-copy-of-training-data), Miller Johnson (https://millerjohnson.com/ninth-circuit-draws-dmca-line-for-ai-generated-code/), Weintraub Tobin briefing (https://www.weintraub.com/2026/10/the-briefing-copied-generated-and-forfeited-the-ai-copyright-ruling-in-doe-v-github/)

Read Next