AI Models Can Use Books. Anthropic Wins Court Victory

AI models can use books, even copyrighted ones, to train. There is no copyright infringement, at least in the US: that is the gist of a somewhat revolutionary ruling by a US federal judge this week that Anthropic's artificial intelligence did not violate the law when it used copyrighted books to train its chatbot Claude, without the consent of the authors or publishers of the texts.

The decision, handed down Monday by Judge William Alsup of the U.S. District Court for the Northern District of California, is seen by experts around the world as a victory for AI companies, which have faced copyright lawsuits from writers and news organizations for using their work to train AI systems.

It is true that the ruling does not apply to Europe, where more restrictive rules apply; but the very clear direction that jurisprudence is taking in the US can still weigh on the European balance of power between publishers, authors on the one hand and AI companies on the other. If Europe remains isolated in the protection of copyright against AI, it will be more difficult for it to maintain its point, also in light of the geopolitical relations between the US and the EU.

The reasons for the ruling are striking. Alsup stated that training with books is a transformative use, it is not plagiarism (it falls under the American “fair use” discipline) and it is similar to an aspiring writer who reads copyrighted texts “not to replicate or supplant” those works, “but to create something different”.

The ruling concerns a lawsuit filed last year against Anthropic by three authors — Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson — who allege that the company used their work without their consent.