Key points
- Part 3 of the US Copyright Office AI report addresses whether training generative AI on copyrighted works constitutes fair use.
- The report concludes that fair use turns on the specific facts of each case; some training uses are likely fair, others are not, with market effects central to the analysis.
- The report rejects the argument that AI training is inherently non-expressive, holding that models generating expressive content cannot categorically claim fair use.
Summary
Released in pre-publication form on 9 May 2025, Part 3 of the US Copyright Office's multi-part AI report analyses the fair use question for generative AI training on copyrighted works. The Copyright Office states that no substantive changes will be made in the final version, making the pre-publication version the operative text. The report's case-by-case framework and rejection of categorical fair use claims for AI training are the operative US analysis for anyone working on training-data questions. It is the reference document for the US dimension of the training-data debate that animation programs cover in IP and ethics curricula.
Implications for animation education
Animation programs in the US teaching AI tools that involve training on existing animation, illustration or visual arts works need to understand that the US fair use question is not settled by a single rule. The Copyright Office's case-by-case framework means that different training scenarios, using public domain works versus commercially licensed works versus scraped web content, may attract different legal outcomes. Programs covering AI ethics and IP should present the report's four fair use factors analysis, particularly the weight given to market effects, as the operative analytical framework rather than any categorical rule. The rejection of the inherently non-expressive argument is significant for student understanding: tools that generate expressive visual content are not insulated from copyright scrutiny on the training side.
Related items
- Copyright and Artificial Intelligence, Part 1: Digital Replicas
- Copyright and Artificial Intelligence, Part 2: Copyrightability
- Getty Images (US) Inc v. Stability AI Ltd: UK High Court Judgment
Source
Source: US Copyright Office ↗ (Policy Text)
Primary document: copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-3-Generative-AI-Training-Report-Pre-Publication-Version.pdf ↗
Cite this item
US Copyright Office (2025). ‘US Copyright Office Report on Copyright and AI, Part 3: Generative AI Training’, US Copyright Office. Available at: https://www.copyright.gov/policy/artificial-intelligence/
Your reference manager can also read this page directly: with the Zotero (or Mendeley) browser connector installed, save it straight to your library. Whole-collection exports: RIS, BibTeX, CSL-JSON.