Generative AI tools are trained on massive collections of public and copyrighted materials. These materials are scraped from the internet without the creators' consent and with no opportunity for recourse other than lawsuits. As of winter 2025, several lawsuits are being processed related to the use of copyrighted works in training AI—the plaintiffs claim that using copyrighted works without permission infringes copyright. In opposition, some legal scholars have pointed out that non-consumptive uses of copyrighted content have been considered fair use in previous court cases (Google Books, HathiTrust).
While these cases are in process, we won’t have definitive answers about whether the use of copyrighted works in AI training data is legal.
In contrast, the US Copyright Office and courts have stated clearly that because human authorship is required for copyright protection of a work, generative AI generally does not create copyrightable output.
Aside from the question of copyrightability, it is also important to understand that output created by generative AI tools is not guaranteed to be accurate or free from legal liability. It is well documented that generative AI sometimes creates “hallucinations,” which are outputs that purport to be true but are demonstrably false. Additionally, textual or visual outputs created by generative AI might subject a user to liability such as copyright infringement. This becomes more likely when you are using, for instance, large chunks of code from generative AI tools or are asking a generative AI tool to create an image of a well known entity like Snoopy or Mickey Mouse.