Part 2 of the government’s report on copyright and artificial intelligence released last month, and it addresses whether people can claim copyright of materials they create using generative AI (e.g., tools like ChatGPT or Sudowrite).
For years now, the US Copyright Office (USCO) has dealt with copyright filings for materials that are in part or wholly generated by AI, and so far Part 2 reaffirms the positions it has already taken as part of existing cases. (Here is one such example.) Even so, you’ll find interpretations of this report that, at first glance, appear to be in conflict from the headline alone. Consider:
- US Copyright Office Changes Stance on AI-Assisted Artwork (Straight Arrow News)
- US Copyright Office Says AI-Generated Content Can Be Copyrighted—If a Human Contributes to or Edits It (VentureBeat)
- Copyright Office AI Report Says Creative Prompting Doesn’t Constitute Authorship (IPWatchdog)
- The US Copyright Office Declares That “Use of AI to Assist in the Creative Process Is Copyrightable” (Gigazine)
- Copyrighting AI-Generated Works Requires “Sufficient Human Control Over the Expressive Elements”—Prompts Are Not Enough (The National Law Review)
And so on. Here are some of the word-for-word conclusions and recommendations of the US Copyright Office (emphasis mine).
- “Questions of copyrightability and AI can be resolved pursuant to existing law, without the need for legislative change.” In other words, the USCO believes existing copyright law remains adequate in a world with AI use.
- “Whether human contributions to AI-generated outputs are sufficient to constitute authorship must be analyzed on a case-by-case basis.” This is comparable to how fair-use cases are evaluated. There is no specific word count that allows you to claim fair use; each and every instance of quoting or using someone else’s work must be assessed on a case-by-case basis by applying fair-use doctrine.
- “Human authors are entitled to copyright in their works of authorship that are perceptible in AI-generated outputs, as well as the creative selection, coordination, or arrangement of material in the outputs, or creative modifications of the outputs.” If you use AI to write your work, you can own copyright for creatively manipulating and arranging the outputs, but not copyright over every single line and paragraph that’s AI generated.
The big questions this raises for me: What is “sufficient” for human authorship? And how do you determine what is human in a work produced with the assistance of AI? There is no tool today that can tell with 100 percent accuracy what is AI generated versus human generated.
Publishing consultant Bill Rosenblatt wrote on a private listserv, “This all means that (assuming Congress takes the USCO’s recommendation and doesn’t pursue legislation in this area) these issues will go to courts to decide. This means we’re in for a good solid decade of uncertainty, at a minimum. Here’s why this is a bigger deal than you might think. Defense lawyers in copyright infringement suits always try to challenge the plaintiff’s copyright registration, saying that some i was not dotted or t crossed properly. Now we’re going to see them do this on the basis that some aspect of the work was ‘AI-generated’ and therefore shouldn’t have gotten copyright protection. E.g., they find evidence that some photo was generated using an ‘AI-powered’ Photoshop feature or plug-in (given that ‘AI-powered’ is essentially a marketing term whose definition is a moving target). And the definition of ‘sufficient’ human involvement is going to be argued about ad nauseam. Judges will have to decide these on, as the USCO says, a case-by-case basis. This will beget a generation of court decisions that lawyers can cite as caselaw for when they are trying to invalidate or defend a copyright registration. And the cycle will go on. Just like it has done for fair use over the past 20 years. It’s gonna be a fun ride.”
The Authors Guild has released their own statement, saying that the Copyright Office’s guidance aligns with their position on AI. They make the following points:
- Using AI to research or to generate outlines or brainstorm does not have to be disclosed or disclaimed. Using AI for such purposes does not affect one’s copyright. Moreover, asking the AI to change your own work would not affect the underlying copyright. They write, “If you wrote a story then put it into a chatbot and asked it to change two of the characters, you would still own the copyright in the original story and could prevent anyone from using all of the elements that carried over in the AI-revised story.” But any expression added by the AI would not be copyrightable. (How is that determined? Unclear unless authors are scrupulously tracking every word and scrupulously honest.)
- AI-generated work in an otherwise human work does not affect copyright in the work as a whole. Or, if a work is a compilation of AI-generated text, the work is eligible for copyright as a compilation, but the AI-generated text itself is not protected.
- Existing law offers precedents for dealing with copyright cases where some portions of the work are not protectable. For example, software copyright cases are one area where judges have learned how to “filter” unprotected elements and only compare the protected parts, according to Brandon Butler, executive director of the Re:Create Coalition.
The Copyright Office and the Authors Guild directly state that prompting alone, even using complex and multiple prompts, is not sufficient to gain copyright protection over the output. Even so, given the sophisticated use of AI by some writers, situations may not be clear cut. In my interview with Elizabeth Ann West last year, she described four different authorial styles of using AI to brainstorm and write: gardener, baker, weaver, architect. Bakers “bash together” the outputs of various prompts and models to create their work, with varied amounts of editing. I think it’s fair to say no one is measuring how much of the result has been touched by human hands. How much does that matter in the end, if—for all intents and purposes—there is zero chance another person could generate the same result?
The Authors Guild admits, “Establishing as a factual matter whether any particular element of a work is AI generated … may present obstacles in infringement cases, making discovery even more laborious than it already is. For instance, an author may not remember which sentences, paragraphs, or elements were AI generated, or they may have died and not kept records. For now, there is no good human or mechanical way to distinguish human-authored from AI-generated text on a word-by-word or sentence-by-sentence basis. AI-generated stories give themselves away by their style, their odd narratives, and sometimes their nonsense, but it is not so obvious if a sentence here or there is AI generated.”
Bottom line: As someone who regularly fields questions from writers on copyright—what requires permission, what constitutes fair use, and what’s protected under copyright law—I believe most writers will remain unsure what’s eligible for copyright if they use tools like ChatGPT. When the AI tools are automatically embedded everywhere, as they now are in Word and Google Docs, the picture will only get muddier. Writer and industry observer Josh Bernoff writes, “What the report fails to address is that AI will be present in everything, in ways that go far beyond simple prompting. … This means every piece of content generated in the future will be a collaboration between human and machine, in a way that makes it very difficult to tease apart who is responsible for the creativity.”
More broadly, while attending a global AI summit in Paris, Kevin Roose of the New York Times wrote, “At times this week, listening to policymakers discuss how to govern AI systems that are already several years old—using regulations that are likely to be outdated soon after they’re written—I’ve been struck by how different these time scales are. It feels, at times, like watching policymakers on horseback, struggling to install seatbelts on a passing Lamborghini.”
Jane Friedman has spent her entire career working in the publishing industry, with a focus on business reporting and author education. Established in 2015, her newsletter The Bottom Line provides nuanced market intelligence to thousands of authors and industry professionals; in 2023, she was named Publishing Commentator of the Year by Digital Book World.
Jane’s expertise regularly features in major media outlets such as The New York Times, The Atlantic, NPR, The Today Show, Wired, The Guardian, Fox News, and BBC. Her book, The Business of Being a Writer, Second Edition (The University of Chicago Press), is used as a classroom text by many writing and publishing degree programs. She reaches thousands through speaking engagements and workshops at diverse venues worldwide, including NYU’s Advanced Publishing Institute, Frankfurt Book Fair, and numerous MFA programs.
