OpenAI and Microsoft Copyright Battle Draws Wave of Amici Briefs from Publishers and Tech Rivals

The copyright dispute between OpenAI and Microsoft has attracted a significant wave of amici curiae — "friend of the court" — briefs from publishers, competing AI companies, and academic institutions, signaling that the case is widely viewed as a precedent-setting moment for AI training data rights. According to Law360's reporting, the volume and breadth of outside interest reflects industry-wide anxiety about a single question: can an AI company train its models on copyrighted material under fair use, or does doing so require a license?

The answer will reshape how every major foundation model is built and priced.

What the Case Is About

The dispute centers on Microsoft's use of OpenAI models — and OpenAI's training practices — in a copyright infringement action brought by multiple publishers and, separately, by individual creators. The consolidated proceedings have become a proxy battle for the entire AI industry because OpenAI's training corpus — like those of Google DeepMind, Anthropic, Meta, and Mistral — includes substantial quantities of copyrighted text, code, images, and other protected works.

OpenAI and Microsoft argue that training on copyrighted data constitutes transformative use under Section 107 of the Copyright Act — the same fair use doctrine that courts have applied to Google Books, image thumbnail search, and other large-scale data processing cases. Plaintiffs argue that AI training is a commercial use that substitutes for, rather than transforms, the original creative works — and that the market harm is material and ongoing.

Why the Amici Wave Matters

Amici briefs are filed by parties who are not litigants but who have a stake in how the court rules. A large amici filing record usually indicates that the case is expected to set durable precedent rather than turn on its specific facts.

The filings in this case are notable for the diversity of interests:

Publishers and content creators — the New York Times, book publishers, and music rights organizations — argue for a narrow fair use reading that would require AI companies to license training data. Their brief argues that the "market harm" prong of fair use analysis strongly favors copyright holders because AI outputs compete directly with the licensed content they sell.

Competing AI companies — including several smaller open-source model developers — filed a brief arguing that mandatory licensing would entrench large incumbents and effectively shut out smaller players who cannot negotiate or pay for the data volumes required to train competitive models.

Academic institutions — MIT, Stanford, and a coalition of university research centers — filed a brief arguing that research and educational uses of AI trained on copyrighted data should be treated distinctly from commercial deployments, and urging the court to distinguish between training a model (transformative) and deploying it commercially (potentially infringing).

The Stakes for Every AI Company

A ruling that training on copyrighted material is not fair use would be the most consequential legal event in AI since the industry's modern era began. The implications cascade through the entire supply chain:

Foundation model developers would face retroactive licensing liability for models already deployed and prospective licensing costs for future training runs. For large models trained on trillions of tokens, licensing every source at scale may be economically infeasible — meaning the industry would either need to shift to synthetic data (AI-generated training data, which brings its own quality concerns) or negotiate blanket licensing arrangements similar to what music streaming services have with record labels.

Enterprise customers licensing API access to these models could face indemnification questions: if the underlying model was trained on infringing data, are the outputs themselves infringing? Most current AI vendor contracts disclaim this liability, but a court ruling could change the commercial risk calculus.

Content creators stand to gain a revenue stream they currently don't receive. The music industry's experience negotiating streaming royalties suggests that licensing AI training is achievable at scale — but the early negotiations were contentious, and smaller creators typically fared worse than major label artists.

Where the Law Currently Stands

U.S. courts have not yet issued a definitive ruling on AI training and fair use. The closest precedent — the Authors Guild v. Google case, in which the Second Circuit ruled that Google's Books scanning project was transformative fair use — is the foundation of AI companies' fair use defense. Plaintiffs argue the analogy is imperfect because Google Books created a search index that directed users to purchase books, while AI systems ingest the content and generate competing outputs.

The Ninth Circuit, where much of the AI litigation is consolidated, has not previously addressed the specific question of whether training a commercial AI model on copyrighted text is transformative. This case may produce that circuit's first authoritative statement.

What to Watch

The volume of amici briefs signals that no party expects a quick settlement. Both OpenAI and the plaintiffs have incentives to litigate to a definitive ruling: OpenAI because a settlement would imply liability and invite further claims; the publishers because a court ruling establishing a licensing requirement would be more durable than any individual settlement.

The first significant procedural milestone is the court's ruling on summary judgment motions — expected in early 2027. If the court denies summary judgment and the case proceeds to trial, the trial itself would likely generate enough testimony and evidence to effectively write the industry's training data policy regardless of the final verdict.