Federal judges have ruled: copying books to feed a language model falls under fair use, that exception which allows using a protected work without authorization for “transformative” use. This decision reshuffles the cards of the global creative economy just as Europe is heading in the opposite direction with the AI Act.
The legal battle opposing writers and AI developers since 2023 has just experienced a decisive turning point with rulings in Bartz v. Anthropic and Kadrey v. Meta. The courts validated the use of copyrighted content for training, provided it does not come from pirate sites like Library Genesis. A precedent that transforms intellectual property into an accessible resource for algorithms, at substantial cost.
The Essentials
- American federal courts qualified AI training on protected works as transformative use falling under fair use in the cases Bartz v. Anthropic and Kadrey v. Meta
- The Supreme Court refused in March 2026 to examine the question of non-human authorship in Thaler v. Perlmutter, leaving current law unchanged
- Europe adopts the opposite approach with the AI Act, which imposes mandatory labeling of content used for training
- The legal battle shifts toward model outputs rather than its training data
Judges Rule in Favor of Technological Innovation
American courts’ reasoning rests on an expansive interpretation of fair use, the doctrine that allows using a protected work without authorization in certain specific cases. The judges determined that training a language model constitutes sufficiently “transformative” use to justify the exception. AI does not reproduce the original work but analyzes it to learn linguistic patterns.
This decision relies on precedents established in the technology sector, notably the 2015 Authors Guild v. Google case where digitizing millions of books for Google Books had been validated. Courts now apply the same logic to artificial intelligence: the commercial use of a technology does not automatically disqualify fair use if the purpose differs substantially from the original work.
The distinction established by judges between legitimate sources and pirate sites reveals an important nuance. Using legally purchased or borrowed works remains protected, but sourcing from Library Genesis or similar platforms exposes one to prosecution. This limit preserves a minimal balance between innovation and respect for copyright.
The Supreme Court Refuses to Rule on Artificial Authorship
The Supreme Court’s refusal to examine the Thaler v. Perlmutter case maintains the status quo on a nonetheless central question: can AI be considered the author of a work? Stephen Thaler had attempted to have his DABUS system recognized as an inventor of a patent and author of artistic creations. Lower courts had rejected his request, ruling that only humans can hold intellectual property rights.
This non-decision by the highest American court leaves economic actors in uncertainty. AI-generated content falls de facto into the public domain, a windfall for companies that can exploit these productions freely. Human creators thus lose a competitive advantage against algorithms capable of producing texts, images, and music at scale.
The absence of clear legal framework benefits tech giants who accumulate AI-generated content without risk of future claims. This situation contrasts with European debates where the question of attribution and traceability of artificial creations already structures regulatory discussions.
Europe Chooses Mandatory Transparency
The European Union takes a radically different path with the AI Act, which has been entering into progressive application since 2025. The regulation requires developers of generative AI models to publish a detailed summary of training data used, including copyrighted works. This transparency obligation aims to enable rights holders to assert their rights.
The European approach favors information and negotiation rather than general exemption. Creators can identify use of their works and demand compensation or removal. This fundamental difference from the American system creates a complex regulatory patchwork for companies operating on both sides of the Atlantic.
The first effects of this divergence are already being felt. Anthropic and Meta adapt their practices according to jurisdictions, using different datasets for their European and American models. This technical fragmentation reflects the emergence of two incompatible digital governance models.
The European Commission justifies this approach by the protection of creative industries, a sector employing 7.5 million people in the Union. The United States prioritizes technological innovation, believing that productivity gains will ultimately benefit creators themselves.
Creators Reorient Their Strategy
Facing these judicial decisions, American authors’ organizations are abandoning their lawsuits over training to focus on model outputs. The Authors Guild and the Writers Guild of America are exploring new legal avenues, notably identity misappropriation when AI mimics a specific author’s style or unfair competition in commercial productions.
This tactical reorientation relies on recent cases where AI models faithfully reproduced the style of contemporary writers, raising questions of economic parasitism. Courts have proved more receptive to these arguments than to claims about training itself.
In parallel, new economic models are emerging. Platforms like Stability AI offer voluntary licenses to creators wishing to monetize AI use of their works. These direct compensation systems bypass legal uncertainties by creating an organized market between content producers and algorithm developers.
The music industry is experimenting with similar approaches with framework agreements between labels and generative AI companies. These private negotiations are gradually replacing legal battles as the mode of conflict resolution.
Economic Models Are Being Reconfigured
Despite fair use decisions, generative AI costs remain substantial. Companies negotiate paid licenses and face considerable legal settlements—Anthropic recently concluded a 1.5 billion dollar agreement. Development costs for a new language model now include technical aspects, computing power, engineering, but also these legal and licensing fees.
This new situation modifies the balance between actors with technical infrastructure and those controlling content. Google, Microsoft, and Amazon strengthen their position thanks to their computing capacity and legal budgets, while publishers explore new content monetization models.
Regulatory asymmetry between Europe and the United States, however, creates opportunities for European creators. Their works, protected by the AI Act, acquire differential value against more easily exploitable American content. This regulatory premium could modify investment flows in creative industries.
Early signals confirm this trend: Netflix and Disney are strengthening their European production teams to benefit from this enhanced protection. Europe is betting on this differentiation to maintain its competitiveness against American tech giants.
American judicial decisions establish a global precedent that far exceeds the American framework alone. They signal to investors and companies that technological innovation will take precedence over traditional copyright protection in the coming years. This legal orientation likely accelerates generative AI development while forcing creators to completely rethink their economic models. The divergence with Europe opens a period of regulatory experimentation whose results will determine the global balance between technological innovation and creative protection.