US Copyright Office publishes report on generative AI training
The US Copyright Office has released a pre-publication version of Part 3 of its AI report, on the use of copyright works to train generative AI.
On 9 May 2025 the US Copyright Office released a pre-publication version of Part 3 of its report on copyright and artificial intelligence, covering generative AI training. Part 1, on digital replicas, appeared in July 2024 and Part 2, on the copyrightability of AI outputs, in January 2025. The Office said a final version would follow without substantive changes expected to the analysis or conclusions.
The report concludes that several stages of building generative AI systems involve copying that engages copyright owners' exclusive rights, and that the central question is whether fair use applies. It says this depends on the facts of each case. Some training uses are likely to be transformative, but commercial use of large quantities of works to produce content that competes with them in existing markets, particularly where the works were obtained unlawfully, goes beyond established fair use boundaries in the Office's view.
On remedies, the Office found that voluntary licensing, individual and collective, is developing in some sectors and considered government intervention premature. It suggested that approaches such as extended collective licensing could be considered where gaps remain.
Why it matters: the report is not binding on US courts, but it is an influential reading of US law at a time when several lawsuits by authors and publishers are under way. For writers and small publishers outside the US, it indicates that the source of training material and the effect on the market for their books are likely to carry weight.
Sources: US Copyright Office: Copyright and Artificial Intelligence; US Copyright Office: Part 3, Generative AI Training (pre-publication version, PDF).