Background
Programmers sued GitHub, Microsoft, and OpenAI over Copilot, alleging that the coding assistant was trained on public repositories and could generate code without the copyright notices and other copyright management information (CMI) attached to the originals. The district court dismissed part of the claim under section 1202(b) of the Digital Millennium Copyright Act, and the Ninth Circuit accepted an interlocutory appeal.
The plaintiffs advanced theories that the system removed CMI during training or distribution and that future users could obtain matching code stripped of identifying information. The appeal required the court to decide how closely an output must correspond to a protected work for the DMCA’s CMI-removal provisions to apply.
The Court’s Holding
The Ninth Circuit affirmed dismissal. Section 1202(b) requires removal or alteration of CMI from a copy of the plaintiff’s work; it does not create a general claim whenever a system produces similar material without attribution. The complaint did not plausibly allege that defendants distributed copies of the named plaintiffs’ code from which CMI had been removed.
The panel also found the asserted future risk too speculative on the allegations presented. At later stages, a plaintiff pursuing prospective relief would need evidence of a substantial risk that the system will reproduce the plaintiff’s code in the legally relevant manner. The court declined to turn ordinary infringement allegations into broader DMCA metadata claims.
Key Takeaways
- A section 1202(b) claim requires a nexus between removed CMI and a copy of the plaintiff’s actual work.
- Similarity or unattributed output is not automatically CMI removal under the DMCA.
- AI plaintiffs seeking injunctive relief must plausibly show a substantial, non-speculative risk involving their own works.
- Copyright infringement and DMCA information-removal claims have distinct elements and should be pleaded separately.
Why It Matters
The decision gives California technology companies and copyright owners important guidance for AI-training and output disputes. It narrows one route to DMCA liability without resolving every possible infringement or attribution theory involving generative systems.
Developers documenting a claim should preserve concrete examples tying outputs to identified works and their original notices. AI providers should continue tracking provenance and output controls, because stronger allegations involving verbatim or near-verbatim copies could present a different case.
For rights holders, the pleading lesson is to identify the work, the CMI originally conveyed with it, the allegedly copied output, and the mechanism by which the information was removed or altered. General allegations about model training or unattributed output may not connect those elements. Providers should keep records concerning training inputs, filtering, output similarity, and notice handling, both to reduce risk and to answer concrete claims. The ruling addresses the DMCA theory certified for appeal; it should not be read as a broad declaration that training, reproduction, contractual, or open-source-license issues can never support relief.
The procedural posture matters when using this opinion. A published Ninth Circuit decision supplies binding circuit law unless later rehearing or Supreme Court review changes it; an unpublished memorandum is generally nonprecedential but still illustrates how the panel evaluated the record. Counsel should separate the rule of law from facts that merely explain the outcome, confirm whether an issue was preserved, and check the mandate and subsequent history before relying on the result. Transactional and compliance teams can use the decision prospectively by documenting the facts that the court treated as decisive. Litigators should build that documentation into declarations, discovery, and the statement of undisputed facts rather than waiting until appeal. Where the panel remanded, the prevailing party still has work to do and should not describe the opinion as a final merits victory. Where it affirmed, lawyers should identify whether the holding rests on statutory interpretation, contract language, evidentiary failure, or harmless error, because that distinction determines how readily the analysis transfers to another dispute.
Read the full opinion (PDF) · Court docket