Generative AI and Copyrightability: Report From the US Copyright Office

In 2023, the US Copyright Office launched an initiative to examine the legal and policy implications for copyright of generative AI (GAI). Incorporating comments from the public, the resulting report will be issued in installments, each addressing a different topic.

Today, I’m discussing the second installment, published in January, which offers the Office’s determinations on what does, and doesn’t, make a work that incorporates GAI eligible for copyright protection. (Part 1 came out in 2024 and covered digital replicas–deepfakes, etc.) There’s a lot of confusion in this area, and while the report does deliver some welcome clarity, there’s still plenty of ambiguity for individual creators and individual works.

Bottom line: works that incorporate GAI may be copyrightable. But there are limitations, and the determination must be made on a case by case basis.

What’s copyrightable?

The existing legal framework, the Office contends, is clear: “No court has recognized copyright in material created by non-humans, and those that have spoken on this issue have rejected the possibility.” For example, the famous monkey selfie case, and a 2023 court ruling upholding the Office’s determination that an autonomously-created (i.e., no human intervention) GAI image was not copyrightable.

Applying this legal framework, the Office comes up with a number of determinations.

Copyrightability may apply to a work where GAI is used as assistive technology. The examples provided mostly draw from the film and music industries: de-aging actors in films, identifying chord progressions in composing. For writers, assistive use of GAI might involve brainstorming ideas, refining plot, assisting with editing, or using speech-to-text applications. Works so created are copyrightable–still, “[w]hile assistive uses that enhance human expression do not limit copyright protection, uses where an AI system makes expressive choices require further analysis.”

Prompts alone don’t provide sufficient control over output to make the GAI user the author of the generated content. Even highly detailed prompts produce different outputs each time they’re used (the Office provides an example on page 19 of the report). Copyright is not about ideas, but about their expression, and “[t]he gaps between prompts and resulting outputs demonstrate that the user lacks control over the conversion of their ideas into fixed expression, and the system is largely responsible for determining the expressive elements in the output.” In other words, too much machine authorship. You cannot, therefore, claim copyrightability on the basis of the prompts you use to create the work–no matter how original you think the prompts are.

Expressive GAI inputs may be partially copyrightable–but it’s case by case. The Office provides an example on page 23: a line drawing used as input, along with a prompt, to produce a fully-rendered image. The Office granted copyright registration to this work–though they split the difference by limiting this to “unaltered human pictorial authorship that is clearly perceptible”, i.e., the degree to which the original line drawing is apparent in the finished output. The Office compares copyrightability here to a derivative work, where protection is “limited to the material added by the later author.”

Works created by human arranging or modification of GAI-generated content are copyrightable. Humans, the Office concludes, “may select or arrange AI-generated material in a sufficiently creative way that the resulting work as a whole constitutes an original work of authorship.” They may also “modify material originally generated by AI technology to such a degree that the modifications meet the standard for copyright protection.” The Office says it has registered several such works, including a comic book combining GAI-created images and human-created text. Again, though, these are case-by-case determinations–which doesn’t seem especially sustainable as a long-term strategy, assuming an increasing volume of such works.

Do the laws need to be changed?

Does copyright law need to be amended to clarify, and perhaps provide additional protection for, GAI-generated material?

The Office summarizes the arguments put forward in support of new legislation, including the contention that protecting GAI-generated material would create new incentives for authorship; that it would empower disabled creators; that it would keep pace with international developments; and that it would provide clarity for creators who work with GAI and are concerned about the copyrightability of their work.

In each case, however, the Office sees no need for change. There are already substantial incentives for authorship; and it’s not clear that additional protection would promote the “progress of science and the useful arts” called for in the copyright clause of the US Constitution, given the concerns of creators whose livelihood is threatened by the proliferation of GAI-generated content (the Office also acknowledges GAI companies’ use of stolen material for training). Copyrightability is already available to disabled creators who use GAI as an assistive tool. As for keeping up with international developments, changing the law is premature: how other countries will apply copyright laws to this rapidly-developing field is very much an open question. Clarity too is a moving target, especially as courts continue to weigh in.

The Office’s conclusion: overall and for the moment, existing copyright law is adequate to determining the copyrightability of works incorporating GAI.

Copyright law has long adapted to new technology and can enable case-by-case determinations as to whether AI-generated outputs reflect sufficient human contribution to warrant copyright protection. As described above, in many circumstances these outputs will be copyrightable in whole or in part—where AI is used as a tool, and where a human has been able to determine the expressive elements they contain….he Office continues to monitor technological and legal developments to evaluate any need for a different approach.

More on GAI and copyright from Writer Beware

Michael Capobianco’s Copyright, Contracts, and AI-Generated Material takes a look at the Copyright Office’s initial effort in 2023 to address the copyrightability of works incorporating GAI, and the implications of its guidance for publishing contracts.

In response to the Copyright Office’s request for public comment, SFWA offered detailed recommendations, including requiring that any GAI-generated material must provably have been based on licensed training data in order to be included in a copyright registration.

Changing laws, not for the better: my post on how the UK’s proposed copyright exception that would allow GAI companies to train on creative material without creators’ permission turns copyright on its head.

My post on several (largely silly, in my opinion) efforts to “certify” human authorship in the age of GAI and GAI haters. None of this, of course, carries any legal weight.