The GitHub AI Training Ruling Separates DMCA Questions From Open-Source License Questions
A September 16, 2026 U.S. appeals ruling narrowed part of a developer lawsuit against OpenAI and Microsoft, but left separate questions about open-source licensing and AI training unresolved.
The September 16, 2026 appellate ruling involving GitHub, OpenAI, and Microsoft narrows one legal theory without resolving the broader question of how open-source software may be used in AI training. Reuters reports that the Ninth Circuit agreed with a lower court's reasoning on the DMCA claim while leaving separate licensing allegations for further litigation. Reuters
One lawsuit can contain several different legal questions
AI training disputes are often described as one giant "copyright question."
They are not.
The developer case discussed by Reuters included a theory under the Digital Millennium Copyright Act involving removal of copyright-management information.
The appellate court's ruling on that theory does not automatically decide whether other copyright or software-license obligations were satisfied.
Why the distinction matters to developers
Open-source software licenses can contain conditions around attribution, notices, redistribution, and other uses.
Those obligations are conceptually different from a claim that a model removed metadata or copied material in a legally actionable way.
A legal ruling on one claim therefore does not create a blanket rule for every use of public code in AI systems.
The unresolved layer is more operational
For developers and AI vendors, the practical question becomes provenance.
Can a company explain:
- where the training data came from?
- which licenses applied?
- what transformations occurred?
- what notices were preserved?
- what outputs may be subject to downstream obligations?
Those are increasingly software-supply-chain questions.
What this ruling does not prove
It does not establish that AI training on open-source code is generally lawful or unlawful.
It resolves part of one case.
The remaining claims, future cases, and different jurisdictions can produce different outcomes.
Observatory interpretation
The durable signal is that AI training law is fragmenting into specific technical and legal questions instead of one universal test.
That creates pressure for better provenance systems and clearer license-aware data pipelines.
Related observation
The issue connects directly to Public Package Registries Are Becoming Enterprise Infrastructure: the software supply chain now includes not only publication and distribution, but potential model-training use.
Sources
Evidence
Sources & further reading
Primary sources, official disclosures, and external research used to ground this report.
- Reuters — OpenAI, Microsoft fend off part of software developer lawsuitreuters.com
Independent September 16, 2026 report on the Ninth Circuit ruling and the remaining open-source licensing claims.
- U.S. Court of Appeals — Doe v. GitHub Inc.cdn.ca9.uscourts.gov
Primary court-record source for the appellate decision and docket materials; exact opinion retrieval should follow the Ninth Circuit docket.
Keep Exploring
Related observations.
AI Crawling Is Becoming a Policy Layer, Not Just a robots.txt Setting
The important change is not another AI crawler blocklist. Search, model training, and user-directed agents are increasingly being treated as different forms of web access with different permissions.
GitHub Actions Is Turning pull_request_target Into an Explicit Security Exception
GitHub is moving a risky GitHub Actions trigger from an easy-to-misconfigure workflow choice toward an explicit policy decision. The change is important for open-source maintainers, especially repositories that process fork pull requests.
CrowdSec's TanStack Disclosure Shows Why Supply-Chain Response Must Survive the First Compromise
The new fact is not that TanStack was compromised in May; it is that a credential stolen then was later used to copy CrowdSec's private repositories, showing why offboarding and token revocation remain part of supply-chain incident response.