For nearly three years, a putative class of software developers has argued that GitHub Copilot — the AI coding assistant built by GitHub and powered by OpenAI’s models — was trained on their copyrighted code and reproduces it without credit or license. On September 16, 2026, the Ninth Circuit Court of Appeals delivered the first circuit-level ruling anywhere in the country addressing how the Digital Millennium Copyright Act applies to a generative AI tool — and it went against the developers.
What the court actually decided
The case, formally Doe v. GitHub, Inc., sought roughly $9 billion in damages on behalf of coders who alleged Copilot’s suggestions sometimes reproduced their licensed code without the attribution or license terms required under open-source licenses. According to reporting from Courthouse News Service, the Ninth Circuit affirmed dismissal of the coders’ core claims, siding with GitHub and OpenAI.
Legal tech outlet ppc.land frames the outcome plainly in its headline: GitHub and OpenAI won the Ninth Circuit appeal over the $9 billion Copilot code claim. And Orrick, counsel in the case, described it in its own announcement as the nation’s first appellate decision on generative AI and Digital Millennium Copyright Act liability — a framing worth taking seriously given how few generative-AI copyright cases have made it past the district court level at all.
The legal theory that mattered — and the one that didn’t
One of the more interesting wrinkles in this ruling, according to a detailed analysis from IP and open-source commentator Heather Meeker, is why the developers lost. Her post, Landmark Decision in Doe v. GitHub AI Court Battle, and a companion breakdown from Quentir explain that the Ninth Circuit held Copilot’s output constitutes “new works” for purposes of DMCA Section 1202(b) — the provision governing removal of copyright management information — and separately found that the plaintiffs’ broader training-data theory had been forfeited procedurally before it was ever properly argued on appeal.
That second point matters as much as the first. It means this ruling may say less about whether training AI models on copyrighted code is lawful in general, and more about how narrowly the DMCA’s specific attribution-stripping provision applies to AI-generated output — and about the importance of preserving every legal theory at the trial court level before appealing. BakerHostetler, counsel tracking the litigation, maintains an ongoing case page at The Copilot Litigation for readers who want to follow how the case (and any further appeal) develops.
How this fits into the broader AI copyright landscape
This decision doesn’t arrive in isolation. Industry newsletter BananaIP’s AI & IP Weekly groups the Copilot ruling alongside other AI-and-intellectual-property developments from the same week, and Norton Rose Fulbright’s ongoing AI in litigation series tracks the broader wave of AI copyright suits working through the courts in 2026 — including the high-profile New York Times v. OpenAI dispute, which Axios reports is heating up on a separate track entirely.
It’s also worth noting that the Ninth Circuit has been active on adjacent copyright-and-technology questions recently; a March 2026 ruling summarized by US Patent addressed when an internet service provider can be held liable for known infringement by its users — a reminder that appellate courts are being asked to apply decades-old copyright doctrine to a wave of new technology fact patterns, AI chief among them.
Why this matters beyond the parties involved
- For AI companies, this is the first real appellate guidance on DMCA exposure for generative output — a meaningfully lower-risk signal than the uncertainty that has hung over the industry since these suits were filed.
- For open-source developers, it’s a disappointing but clarifying result: the specific DMCA theory used here didn’t work, though it leaves the door open for differently framed claims (including the training-data theory that was never properly reached on the merits).
- For copyright and technology lawyers generally, the forfeiture issue is arguably the more important takeaway — a reminder that in fast-moving AI litigation, preserving every theory for appeal is as important as winning it at trial.
The bottom line
Doe v. GitHub won’t be the last word on AI and copyright — the training-data question at its core was never actually decided on the merits — but it’s the first time a federal appeals court has weighed in on generative AI and the DMCA specifically, and it will be cited in every similar case filed from here forward.
This post will be updated if the plaintiffs seek en banc review or petition the Supreme Court.
