Two federal judges in San Francisco issued back-to-back decisions this week that give generative-AI developers an early advantage in the growing wave of copyright litigation. Both courts held that ingesting copyrighted books to train large language models can qualify as “transformative” fair use, although they left the door open to future liability under different facts.
On Monday, U.S. District Judge William Alsup ruled in Bartz v. Anthropic that the Amazon- and Google-backed start-up’s use of lawfully purchased books to train its Claude model is lawful under the Copyright Act. Alsup said the practice is “among the most transformative uses in our lifetime,” but he ordered Anthropic to stand trial over the separate allegation that it stored some seven million pirated titles in a central database, a claim that could expose the company to billions of dollars in damages.
Two days later, U.S. District Judge Vince Chhabria dismissed a lawsuit brought by 13 authors—including Sarah Silverman and Ta-Nehisi Coates—who accused Meta Platforms of copying their works to build the Llama model. Chhabria found the plaintiffs failed to show concrete market harm and said their “wrong arguments” left him no choice but to rule for Meta, while cautioning that training on copyrighted text could still be illegal “in many circumstances.”
The twin rulings are the first substantive U.S. opinions on AI-training fair use and are likely to influence dozens of pending cases against OpenAI, Stability AI and others. Within hours of the Meta decision, a new suit was filed in New York alleging Microsoft used roughly 200,000 pirated books for its Megatron model, underscoring the unsettled legal landscape.
For now, the decisions signal that U.S. courts may view large-scale text ingestion as permissible when the material is obtained lawfully and plaintiffs cannot demonstrate direct market substitution. Content creators, meanwhile, are pressing Congress and regulators to craft clearer rules before more precedent accumulates in the courtroom.
Ruling on the fairness of training generative artificial intelligence with copyrighted works without permission, two federal judges faced the same question: Can machines replace human creators?
Copyright penalties will hurt the United States's ability to stay ahead in its competition with China to develop cutting-edge AI. Congress or the courts should affirm the legality of using publicly available data for training AI models, argues @timhwang and @JoshuaTLevine.
Just in: Meta $META and Anthropic win copyright suits, but rulings come with caveats on AI output and copyright infringement. Anthropic is backed by Amazon $AMZN and Google $GOOGL, reports The Verge. #AI #Copyright
