In its third installment on artificial intelligence and copyright — a pre publication version released on May 9, 2025 — the US Copyright Office addresses the use of copyrighted works in the development and training of generative AI systems. The report makes it clear that "Fair Use" will not always shield AI developers from licensing obligations when training AI systems on copyrighted works.
First, what is Fair Use?
Fair Use is a defense against copyright infringement, which allows the use of copyrighted works without permission from the rights holder if the use meets certain criteria. The doctrine is based on four non-exclusive factors that courts consider to determine whether a particular use is in fact, fair use.
- Purpose and Character of the Use: The courts look at whether the use of the copyrighter material is commercial, nonprofit or other purposes and whether the use is "transformative". Transformative use refers to whether the use of the material creates or adds new expression, meaning, or messaging.
- Nature of the Copyrighted Work: Is the work used for training purposes of a factual or creative nature? Creative works usually receive stronger protection.
- Amount and Substantiality of the Portion Used: his examines how much of the work is used and whether the portion used is significant to the entire work.
- Effect of the Use on the Market: This looks at the impact of the use on the market for the original work and potential market harm.
When it comes to AI, the copyright office applies the four factors as it would to any other infringement issue: Is the use transformative and does it add new expression or meaning? If so, it may be more likely to be considered fair. Was the work used factual or functional or was it highly creative ? Did the use encompass entire works or more limited copying? And finally, the report introduces the idea of market dilution, where AI-generated content could negatively impact the market for original works (such as lost sales, lost licensing opportunities etc...), and significantly affect creators.
How does the concept of Fair Use apply to Artificial Intelligence?
While some uses of copyrighted works for AI training may qualify as fair use, others will not, and each case must be evaluated on its own merits. The courts are still trying to determine whether fair use can be a defense against these claims. Each use case involves balancing the transformative nature of AI training against the potential market harm and the extent of copying. Whether AI training is transformative will also depend on how the resulting AI application is used. For example research-driven applications are more likely to be highly transformative than applications that produce outputs more similar to the copyrighted works.
What should AI Developers do?
Keep in mind that these factors will be applied based on the special circumstances of every individual case in order to both protect the copyright rights of owners and enable innovation and creativity. Developers should assess whether their use of copyrighted work is transformative and justified. Uses that add new expression or serve a different purpose may be more likely to qualify as fair use. Here are several examples of fair use from the report:
- Google Books: The Second Circuit found that Google's scanning of books to create a full-text searchable database was transformative. The purpose was to provide information about the books' contents, which is different from the original purpose of the books themselves.
- Google LLC v. Oracle America, Inc.: The Supreme Court concluded that the "reimplementation" of copied code was transformative because it furthered the development of computer programs by enabling programmers to use their existing skills in a new mobile platform.
- Sega Enterprises Ltd. v. Accolade, Inc.: The Ninth Circuit held that copying computer code to learn the functional requirements for hardware-compatible games was transformative. The purpose was to enable the creation of new, interoperable video games.
Other considerations:
- If you use highly creative work, it may be less likely to be considered fair use compared to factual or functional works.
- While copying may be considered the ultimate form of flattery, copying an entire work will not be deemed fair use unless it is necessary for a transformative purpose.
- Consider the market impact your use will have including market harm, lost sales, market dilution and lost licensing opportunities.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
