- Federal courts are letting artists’ copyright infringement claims against AI companies proceed to discovery, keeping significant liability exposure open for model developers.
- Anthropic’s $1.5 billion settlement in September 2025 over pirated training data now anchors what courts and defendants treat as the cost of unauthorised ingestion, roughly $3,000 per book for affected authors.
- The UK and U.S. courts have reached conflicting conclusions on whether AI model weights constitute infringing copies, leaving cross-border compliance unresolved for developers operating in both markets.
A $1.5 billion settlement, a Supreme Court ruling on AI authorship and an active New York Times case heading into sanctions territory, AI copyright law is no longer abstract. The cases now moving through U.S. and UK courts are drawing practical lines around who owns what when a machine does the creating, and the outcomes will set compliance floors for every developer training on third-party content.
Training Data and Fair Use
The most contested question in AI copyright litigation is whether training on copyrighted works qualifies as fair use under U.S. law. Courts examine four factors: the purpose and character of the use, the nature of the copyrighted work, the amount used and the effect on the market for the original.
In Bartz v. Anthropic PBCa federal court found that training Claude on lawfully purchased print books was highly transformative fair use. The same court drew a hard line at pirated books used in a central training library, finding those uses infringed copyright. A separate ruling in Meta Platforms, Inc. found that training LLaMA on shadow libraries was also “highly transformative,” but only because plaintiffs had not produced sufficient evidence of market harm.
Taken together, these rulings suggest courts may accept the transformative nature of AI training as a starting point, while placing the evidentiary burden for market harm on plaintiffs. How the training data was sourced, purchased, scraped or pirated, remains a separate and decisive question.
The Human Authorship Requirement
While training data disputes dominate the headlines, courts have been consistent on a related question: who, or what, can hold a copyright. The answer, confirmed at every level of the U.S. judiciary, is that only human-created content qualifies.
Thaler v. Perlmutter tested that boundary directly. Stephen Thaler sought copyright protection for artwork generated autonomously by his “Creativity Machine” AI system. The D.C. Circuit affirmed the human authorship requirement in March 2025, and the Supreme Court declined to hear the case, leaving that standard intact. AI-assisted work can qualify for protection where a human exercised meaningful creative direction, but the line between tool and author will keep generating disputes as generation capabilities advance.
Outputs and Market Substitution
The New York Times v. OpenAI and Microsoftfiled in December 2023, sits at the centre of a third front in this litigation wave. The Times alleges that OpenAI trained on its journalism without authorisation and that the resulting products can reproduce articles closely enough to compete directly with the newspaper. In March 2025, Judge Sidney H. Stein denied most of OpenAI’s motions to dismiss, allowing the central infringement claims to proceed. The case has since moved into discovery, with the Times filing a request for sanctions against OpenAI in July 2026 over alleged discovery misconduct concerning training data sets.
Getty Images v. Stability AI illustrates how jurisdiction shapes outcome. In November 2025, the UK High Court largely rejected Getty’s secondary infringement claim, ruling that Stable Diffusion models do not store training data as infringing copies. Getty had already abandoned its primary training claim in the UK for want of proof that training occurred there. In the U.S., Getty refiled in the Northern District of California, where Judge Thompson in 2026 dismissed DMCA claims but allowed direct copyright infringement, trademark and unfair competition claims to continue. The divergence between jurisdictions on what AI model weights actually are, infringing copies or something else entirely, remains unresolved.
Settlements and Licensing
Litigation has not been the only response. The music industry shows both strategies running in parallel. Universal Music Group Sony Music Entertainment and Warner Records, through the RIAA, sued AI music generators Suno and Udio in June 2024 for direct copyright infringement of sound recordings and lyrics, alongside DMCA circumvention claims. By November 2025, Warner Music Group had settled with Suno and formed a licensing partnership; both Universal Music Group and Warner Music Group reached separate licensing agreements with Udio. As of April 2026, Sony Music continues to litigate against both services.
The Bartz v. Anthropic settlement followed a similar logic. Judge Alsup gave preliminary approval in September 2025 to a $1.5 billion class-action settlement covering authors whose pirated books were used in training, with affected authors potentially receiving around $3,000 per book. For rights holders weighing their options, that figure now anchors what courts and defendants treat as the cost of unauthorised ingestion.
Proprietary Data and Cross-Border Complexity
Thomson Reuters v. ROSS Intelligence points toward a stricter standard when training material is commercially valuable and proprietary. The case involved a non-generative AI platform that allegedly copied from Thomson Reuters‘ Westlaw database to build a competing legal research service. The court rejected a fair use defence, a result that carries direct implications for enterprise AI developers sourcing training data from licensed databases rather than the open web.
The jurisdictional dimension adds another layer. The UK High Court’s ruling in Getty v. Stability AI turned partly on whether model weights constitute infringing copies, a question with no settled answer in either jurisdiction. Getty appealed the UK decision in December 2025. How that appeal resolves, alongside the U.S. cases still in discovery, will shape compliance frameworks for AI developers operating across both markets. The EU’s August 2026 transparency deadline adds a parallel compliance pressure for developers with European exposure.
