bbieron@platformeconomyinsights.com

Federal Judge Delivers Ruling That AI Training is Copyright Fair Use

Jul 1, 2025

Report from The Verge

In Brief – Federal Judge William Alsup has ruled on summary judgement that artificial intelligence company Anthropic did not break the law when it used copyrighted books to train its AI chatbot without the consent of the authors or publishers, setting a strong precedent for the application of the “fair use” doctrine to large language model training. However, he sided with the author plaintiffs on charges related to Anthropic’s use of pirated books to build what was described as a massive “central library” held for unnamed future applications, a use that the judge rejected as transformative and failed the test for fair use. Alsup’s strong affirmation of the applicability of the fair use exception to the training of Generative AI large language models (LLMs), which he described as a highly transformative use, represents a win for AI developers who have faced numerous copyright lawsuits from writers, media organizations, image creators and musical artists. However, the decision also hinged on facts that might not be present in other cases, including that Anthropic’s chatbot was not alleged to create copies of the authors’ works, so copying only involved “inputs” for the LLM training, not outputs that violated copyright.

Context – Copyright questions around the legality of training neural networks are probably the biggest legal and regulatory issues surrounding today’s most popular AI services. In the US, lawsuits have center stage and legislation is unlikely, at least before enough judges rule. In the EU, with its AI Actregulators and AI expert groups will play key roles. As noted, in this case, the were no claims that the AI system was creating copies. That is not the case with some image generators or in the New York Times lawsuit against OpenAI. Finally, although Judge Alsup avoided ruling on how the fair use doctrine applies when there is conscious piracy to gather content for LLM training, he warned, “Such piracy of otherwise available copies is inherently, irredeemably infringing even if the pirated copies are immediately used for the transformative use and immediately discarded. But this order need not decide this case on that rule.” With that, copyright holders can consider this a split decision.

View By Monthly
Latest Blog
EU Commission Accepts X’s Plan to Address DSA Transparency Concerns

Report from MediaPost In Brief – The European Commission has announced the approval of X’s plan to address several Digital Services Act (DSA) compliance shortcomings identified by the regulator. The decision follows the €120 million fine imposed on the social media...

TikTok’s Latest EU DSA Charges Allege Failing to Protect Minors

Report from Reuters In Brief – TikTok faces fresh allegations from the European Commission that its platform design violates the EU's Digital Services Act (DSA) by failing to adequately protect minors. The preliminary findings mark the fourth DSA case opened against...

New Jersey Bans So-Called “Surveillance Pricing” for Groceries

Report from NJ.Com In Brief – New Jersey Gov. Mikie Sherrill (D) has signed the Fair Price Protection Act, banning grocery stores and third-party grocery delivery platforms from using consumers' personal data to set or vary food prices. Sherrill said the measure will...

France Officially Sets 15-Year-Old Age Threshold for Social Media

Report from the New York Times In Brief – France has become the first European country to fully approve a nationwide ban on social media for children under 15, with the law expected to take effect as early as Sept. 1. Passed by large majorities in both houses of the...

Platform Economy Insights produces a short email four times a week that reviews two top stories with concise analysis. It is the best way to keep on top of the news you should know. Sign up for this free email here.

* indicates required