Generative AI

Authors are using generative AI to support their creative labors, engage in research across a number of disciplines, and also use it for other mundane but important tasks. But, there are still legal uncertainties when it comes to how authors and creators should interact with Generative AI. We will update this page as well as publish new resources as legislation and case laws develop in this arena.

Frequently Asked Questions

Note to readers: We welcome suggestions for this FAQ page. If you think there’s a question that should be answered here, please email info@authorsalliance.org with the subject line “AI FAQ suggestion.” 

I’m an author who wants to use AI in my work

I’m an author and I’m concerned that my work has been used to train AI

I’m a researcher who is using AI in my work

I’m an author who wants to use AI in my work

If I use AI to generate a work, is that work copyrightable?

This is a fairly complex question and one that will likely become more nuanced as time goes on. We expect guidance on this question to evolve, so be sure to triangulate this information with other information sources.  

First, under U.S. law, an AI-generated work, without any human intervention, is currently considered uncopyrightable. These are the facts of Thaler v. Perlmutter, which confirmed that non-human machines cannot be authors under US copyright law. So, if you are simply submitting a prompt to an AI tool and using the result without any further intervention, that AI generated work is likely uncopyrightable. 

However, this rule against copyright for purely AI-generated works begins to break down when human authors exert control over the final work. If you manually edit an AI-generated work and make copyrightable modifications to the work, then you will typically hold copyright in those modifications. If you select and arrange otherwise uncopyrightable AI-generated works (a set of AI-generated illustrations for a children’s book, for example), it is likely that you will be able to claim copyright in that selection and arrangement.  

Although it is important to remember that the US Copyright Office cannot make conclusive decisions on the copyrightability of a work (only a judge can), the US Copyright Office’s decision to register a work still serves as prima facie evidence regarding the validity of a copyright. 

The US Copyright Office registered the AI-assisted A Single Piece of American Cheese, based on the selection, coordination, and arrangement of AI-generated elements. Working through similar issues related to Zarya of the Dawn, the Copyright Office registered the work’s text and the selection, coordination, and arrangement of the work’s written and visual elements. The Copyright Office refused to register the images contained in the work, which had been generated by Midjourney without further modification from a human author.    

If you want to explore this topic further, we would recommend that you spend some time with the U.S. Copyright Office’s Copyright and Artificial Intelligence resources. For registration advice related to registering works containing AI-generated content, the Copyright Office has produced a useful webinar. 

Finally, be aware that there may be some variation in how different countries approach this question of copyrightability of AI-generated works. Some Chinese courts, for example, appear to have recognized copyright in AI-generated and assisted works. We expect to see an evolving international landscape in the coming years and we’ll be sure to post updates as new details emerge.  

If I am registering a work with the US Copyright Office, but some of it was generated by AI, what should I do? 

The US Copyright Office has provided “Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence.”  

If there are purely AI-generated elements within your work, the Copyright Office requires you to exclude those elements from the registration application (not doing so may result in refusal of registration, as was the case in Allen v. Perlmutter):  

“AI-generated content that is more than de minimis should be explicitly excluded from the application. This may be done in the “Limitation of the Claim” section in the “Other” field, under the “Material Excluded” heading. Applicants should provide a brief description of the AI-generated content, such as by entering “[description of content] generated by artificial intelligence.” Applicants may also provide additional information in the “Note to CO” field in the Standard Application.”

When you register a work with the US Copyright Office, you are asked to fill out a variety of online forms related to the work. During this registration process, you will have an opportunity to exclude the AI generated portions of your work. One of the most helpful resources providing more fine-grained information about this is the US Copyright Office’s Webinar: Registration Guidance for Works Containing AI-generated Content.  

Excluding AI-generated content is not a particularly detailed or complicated process at this time. Mainly, it involves noting that there is some AI-generated content in your work and that you do not intend to claim copyright in that content. The US Copyright Office is not requiring you to clearly delineate all uncopyrightable AI-generated content. While it may be prudent for you to keep detailed records of this information (among other things, you may forget over time), the Copyright Office requires minimal information related to this question. 

Detailed records will also serve you if you ultimately wish to register the selection and arrangement of AI generated works. In the case of A Single Piece of American Cheese, the successful registration of the AI assisted work was based in part, by the submission of this video, detailing how the work was made.

Since AI-generated works lack a human author, are they essentially treated as public domain, meaning anyone can use them without permission?

Yes, a purely AI-generated work, with no human control or intervention, would effectively be in the public domain in the United States. This fact may create some very exciting opportunities over time. For example, we’d love to see collections of useful public domain AI-generated works to meet needs for Open Educational Resources — perhaps a set of accurate and freely available AI-generated medical illustrations that could be incorporated into no-cost resources for medical training as well as used for journal articles.  

Be aware that, in some instances, your use of an otherwise public domain AI-generated work may be encumbered by contractual obligations through negotiated contracts or unilaterally drafted terms of service. These contractual obligations may restrict what you can do with an otherwise public domain work. Monetary damages may be low for breach of contract claims, when compared with copyright infringement claims; but there may be other detrimental nonmonetary consequences—such as being blocked by a vendor or a service. This is not a new problem and not unique to AI-generated works.  

While the work would effectively be in the public domain, there is another quirk to point out — that AI-created work could be subject to the claim that it is substantially similar to an already existing in-copyright work, if it can be proven that the prompt-author or even the training of the AI model utilized the in-copyright work, potentially resulting in copyright infringement. (Substantial similarity is a level of similarity that shows improper appropriation of the plaintiff’s work, one of the requirements for an infringement claim) See our treatment of this subject here.  

Can AI-created works be subject to copyright infringement claims?

Yes, they can. Disney v. Midjourney is one such case. The complaint in that case provides many examples of potentially infringing outputs.

While an AI-generated work, without human intervention, is not copyrightable, it may infringe an already existing work. Under US law, you do not have to intend to infringe another’s work — even “innocent infringers” can be found liable for copyright infringement. If your work is found to be substantially similar to another in-copyright work and you had access to that in-copyright work, then you could be liable for copyright infringement (see also the question on unintentional infringement below). 

Ultimately, a copyright infringement claim related to an AI-created work would be largely the same as a claim against a human-authored work. The fundamental substantial similarity analysis, which compares the two works without reference to the infringer’s state of mind, would remain the same.  

If you are intentionally trying to replicate an in-copyright work with AI (e.g., your prompts are explicitly referring to another in-copyright work and asking the model to replicate that work), that is likely not very hard to accomplish for some types of work. While most AI tools have guardrails around reproducing large quantities of text (e.g., a Harry Potter chapter or a full newspaper article), these guardrails currently appear to be less robust for images, at least among some models. See this complaint against Midjourney.  

Regardless of model, it’s not hard to imagine that one could successfully prompt an AI to generate an image that would be substantially similar to another in-copyright image. Because an infringing substantially similar image need not be identical to the original, it would be quite challenging for AI systems to anticipate and prevent the generation of all potentially infringing works.  

If you are engineering your prompts to deliberately infringe another work, then you may be liable for infringement. This should not be particularly surprising. Although, just as with other potential claims of copyright infringement, you still can legally generate works that are substantially similar if they qualify as fair uses, including parody. As a public policy matter, therefore, we would not like to see a future where AI is constrained in ways that hinder its ability to produce substantially similar works for fair use purposes.

What is the risk that your AI-generated content might unintentionally infringe someone else’s copyrighted material, and who is liable if that happens?

In general, we view the risk of AI-generated content unintentionally infringing someone else’s copyright as fairly low, though this is admittedly new territory with no legal cases to point to at this time. 

It is important to remember that users cannot easily unintentionally infringe the copyright of an existing work, because a copyright infringement claim requires the plaintiff to prove that the infringer had access to their work. If they cannot prove you had access to and copied their work, you cannot be liable for a copyright infringement claim. (Note: In the case of more famous and widely distributed works (e.g. Mickey Mouse), access may be presumed). 

To account for the possibility of unintentional infringement by users, some AI developers have offered limited indemnification for users. A statement for some of Google’s services can be found here. While this indemnification is clearly beneficial to users, we would recommend that you look closely at the AI platform you are using and its associated terms. For example, for OpenAI, If you look at the scope of the indemnification language in its terms of service, you will notice that it is limited in its protection (“This indemnity does not apply where: (i) Customer or Customer’s End Users knew or should have known the Output was infringing or likely to infringe…”). How useful these indemnification clauses may be in the future remains to be seen.

Because access could more easily be demonstrated by investigating the training dataset, AI companies are more likely to be subject to infringement claims, even when they have never intended for their models to generate infringing materials. It’s possible that certain AI models will develop a reputation for readily producing infringing content, regardless of the intention of the user. We would suspect that such models will be very vulnerable to lawsuits — rightsholders will most certainly be on the lookout for this possibility.  

Do AI tool providers (like OpenAI or Midjourney) claim any rights over the content I input into their systems or the output they generate, according to their terms of service?

Terms of Service can change over time, so be careful to verify what we share here.  

In OpenAI’s Terms of Use (accessed July 3, 2025), we find: 

“Your content. You may provide input to the Services (“Input”), and receive output from the Services based on the Input (“Output”). Input and Output are collectively “Content.” 

“Changes to these Terms or our Services. We are continuously working to develop and improve our Services. We may update these Terms or our Services accordingly from time to time.” 

and:

“Ownership of content. As between you and OpenAI, and to the extent permitted by applicable law, you (a) retain your ownership rights in Input and (b) own the Output. We hereby assign to you all our right, title, and interest, if any, in and to Output.”

and: 

 “Our use of content. We may use Content to provide, maintain, develop, and improve our Services, comply with applicable law, enforce our terms and policies, and keep our Services safe.”

“Opt out. If you do not want us to use your Content to train our models, you can opt out by following the instructions in this article. Please note that in some cases this may limit the ability of our Services to better address your specific use case.”

—

These terms mean that OpenAI currently claims no rights in the inputs you might submit into their system or outputs generated by OpenAI services. The terms also inform users that they may choose to opt out of OpenAI using their Content for training AI. However, even if you opt out of AI training, their terms also mean that OpenAI can make any internal use of your Content (including inputs and outputs). It is highly likely that anything you upload or otherwise submit to OpenAI may be used in ways that might not be easy to identify at this time, even if it is not directly training AI. 

—

Midjourney’s Terms of Service (accessed July 3, 2025) are actually far broader, more complicated, and explicitly contemplate a range of possible future uses of your work: 

  1. Midjourney retains ownership of some content created by users: “If you are a company or any employee of a company with more than $1,000,000 USD a year in revenue, you must be subscribed to a “Pro” or “Mega” plan to own Your Assets.”
  2. Midjourney reserves significant rights to your Content (which includes anything you input into the system and anything you output): “By using the Services, You grant to Midjourney, its affiliates, successors, and assigns a perpetual, worldwide, non-exclusive, sublicensable no-charge, royalty-free, irrevocable copyright license to reproduce, prepare derivative works of, publicly display, publicly perform, sublicense, and distribute the Content You input into the Services, as well as any Assets produced by You through the Service. This license survives termination of this Agreement by any party, for any reason.”

In looking through the representations Midjourney requires users to make, the spectrum of possible uses Midjourney may make of its users content or authorize others to make, and the indemnification language in its terms, we would note that users are agreeing, via the terms, to a broad category of future uses, including activities not related at all to the development or improvement of AI. These uses are far more extensive than those contemplated by OpenAI’s current terms. 

To give a specific example, let’s say you generate an image with Midjourney based on your line drawing. Midjourney might use that image in a Super Bowl commercial, with no obligation to compensate you for that use. Further, it might license that image to other companies making commercials, again with no obligation to compensate you. OpenAI’s terms do not currently contemplate any similar use — if OpenAI did the same, you would not have contractually agreed to that use.    

If an AI gives me a story idea or writing prompt, can I use that idea freely without worrying about copyright (since copyright protects expression and not ideas)?

If an AI provides you with an idea or writing prompt (e.g., “Write a five page story comparing unicorns’ migration patterns on different continents”), then you should absolutely not worry about copyright issues. Copyright law in the United States does not protect ideas.  

The principle at work here is that ideas are the fundamental building blocks of creativity. If our copyright law allowed anyone to monopolize these building blocks, our entire system would cease to function. So, yes, again, you should feel free to use basic ideas freely!

Is it legal to use AI to generate new content in the style of a specific author or to create a story using another author’s characters?

We’ve examined the subject of copyrightability and style on our blog. Generally, style alone is not protectable, but specific expressive elements may be copyrightable.  

Characters are frequently copyrightable. The Batmobile was found to be copyrightable. Some famous characters were once restricted by copyright but now free for everyone to use. Sherlock Holmes was subject to copyright but has now fully entered the public domain in the United States (even the later stories).  

Note also that many characters are not copyrightable. For example, in Shelby v. Halicki, the 9th Circuit considered whether “Eleanor,” a collection of Ford Mustangs featured across four films, including Gone in 60 Seconds (2000), was copyrightable. The court held that, “under the Towle test, Eleanor was not entitled to character copyright protection because it (1) did not have conceptual qualities, (2) did not have consistent traits, and (3) was not especially distinctive.” Character copyright can be narrow or nonexistent for characters that are not distinctive and consistent.

I’m an author and I’m concerned that my work has been used to train AI

How do I know if my publisher has licensed my work for AI training or other purposes?

It’s possible that you can find more information on this topic via resources like this Generative AI Licensing Agreement Tracker. 

Even if you don’t find information in these resources, we would recommend that you start by asking your publisher for this information. As we’ve written before, it’s possible that some publishers have licensed works for AI training without consulting authors or even disclosing what materials they have sold to AI companies—many of the AI training data licensing deals are opaque.    

If this is something you’re concerned about in the future, you should examine the language of your publishing agreement closely — while older agreements may not have directly addressed AI training, we would expect that many current and future publishing agreements will feature language related to this possible use.  

I’ve heard of “opt-outs.” How do I do that/can I do that? 

This is a complicated question and evolving territory — that acknowledged, here is one guide to opting out from Wired magazine. If this is a question you’re actively considering, please look for up-to-date resources: given how quickly things are changing, we don’t expect any one resource to be up-to-date for long.  

Here, note that the United States currently has no formal legal requirement regarding opt-outs; in contrast, the EU’s Artificial Intelligence Act gives rightsholders the right to opt out of commercial AI training.  

While methods to opt out are available for many platforms, it is possible that some AI researchers and developers may ignore opt-outs; in a country like the US where there’s no legal requirement for abiding by opt-outs, the incentives to ignore opt outs may be great for some AI developers. 

Keeping your works entirely private is likely to be the only method by which you can ensure that your work will not be used for AI training. The truth is that even a limited physical publication of a work might not prevent AI training — some companies have scanned physical copies and incorporated them into their AI training datasets, so even limiting the publication of your work to physical copies does not ensure that they will not be used to train AI.  

Here, we’d also like to draw your attention to the possible downsides to opting out of AI training. As discussed in this zine related to AI and bias, removing a diverse set of works from the pool used to train AI can amplify bias and other harmful characteristics found in “biased low friction data,” which is typically derived from easily accessible online content. When thinking through whether to opt out, consider the possibility that the inclusion of your work in AI training may help to mitigate this problem.

I’ve published my work under a Creative Commons license. Does that mean it’s being used for AI training?  

Maybe. The fact that you have published your work under a Creative Commons license may make it more attractive to some AI developers, particularly if you’ve applied a broad license to the work, one that permits most types of uses. Here, note that Creative Commons has stated that CC licenses do not permit AI training without attribution.

Many AI developers have made use of works without regard for the work’s copyright status or whether a rightsholder permitted the use because AI training is likely a fair use — for those types of developers, the Creative Commons license applied to your work may neither help nor hinder the use of your work for AI training.  

If you wish to prevent the use of your work in AI training, then researching methods to opt out of AI training, such as utilizing robots.txt, will provide you with your best chances.   

I’ve heard that there are lots of class action lawsuits. How can I tell if my work has been included? Is there a way to opt out of a class? 

Ideally, you should be able to determine whether you are a member of a purported class by reading the description of the purported class in early court documents. By looking at class-related documents within the docket for a specific case, you will find these class descriptions. 

Unfortunately, this is not always the case. At earlier stages of litigation, it can sometimes be difficult to determine with certainty that you are a member of a proposed class. For example, in Bartz v. Anthropic, the plaintiffs’ motion for class certification contained several redactions, which obscured the proposed class members. In other cases, the courts are delaying class certification.

Still, you may have some means of determining whether you are potentially a part of a class. For example, The Atlantic has provided this tool to help users search LibGen, one database that has been used for AI training. Similar search tools exist for other large data sets. While being part of a data set is not equivalent to being a member of a class, it greatly increases the chances.

At more advanced stages in the litigation (once the class has been certified at the very latest), all class members are more transparently defined and will be notified when possible. This notification can come in many forms — you may be notified by physical mail or email, or via an online posting to a specific class-related website. 

If a class is certified by the court AND you are a member of the class, you can opt out. When members of the class are notified, that notification should include procedures to opt out. 

If you believe that you are a member of a certified class and did not receive notification, we recommend that you search for an online notification designed to reach all class members. A straightforward online search will likely lead you to relevant information. Additionally, there are websites dedicated to providing the public with information related to class actions: here is one example and here is another.  

Opting out of the class is one option available to you; in some cases, you might also consider objecting to a proposed class settlement — here is a good example of one such objection.  

I’m a researcher who is using AI in my work.

I am a researcher using copyrighted works with AI to create summaries, synthesize data, or other functions. Can I use those works with AI tools? 

If you’re planning on ingesting/uploading works into an LLM or other type of AI model, there are a number of things you should consider:

  1. If you work for a company or other type of organization, does uploading the work conflict with the policies of your employer? 

Depending on the nature of the works, it’s possible that uploading the works into an LLM may conflict with a policy put in place by your employer. Highly confidential documents are likely a bad fit for uploading to a third party LLM, unless explicitly approved for the use you’re considering and with an LLM with appropriate security protocols. If you are a researcher, it is likely that your institution has policies in place related to this issue. For example, Princeton University has a Generative AI Tools Use Policy along with different data classifications. 

  1. Have you agreed to any terms for using the work you are planning to upload? 

Contractual agreements may limit your use of works beyond what copyright would require. Contracts and terms of use can impose terms on researchers that may prevent your use of the work – if you would like to learn more about this topic, we would recommend that you start with Restricting Innovation: How Publisher Contracts Undermine Scholarly AI Research. 

This is a feature of the law that we would like to see changed – we strongly believe that contractual agreements should not be permitted to undermine fair use rights in the United States. 

  1. What are the terms of use of the AI or LLM that you are using? Do you feel comfortable with uploading works under those terms? 

LLM terms of service will typically require you to represent that you hold all necessary rights in the works that you are uploading. The wording of these terms is friendly to exercises of fair use, but some of the other terms of service might not be as compatible with exercises of fair use. 

For example, Midjourney’s Terms of Service (accessed August 7, 2025), contain this statement of “Rights You Give to Midjourney”: 

“By using the Services, You grant to Midjourney, its affiliates, successors, and assigns a perpetual, worldwide, non-exclusive, sublicensable no-charge, royalty-free, irrevocable copyright license to reproduce, prepare derivative works of, publicly display, publicly perform, sublicense, and distribute the Content You input into the Services, as well as any Assets produced by You through the Service.”

At a minimum, this paragraph seems to presume that you as the user hold the rights that you are agreeing to license to MidJourney (and elsewhere in the terms of service, you represent that you have “ all necessary rights and permissions to provide, edit, generate, and share such content“). If you are not the rights holder in the works you upload into Midjourney, you would not be in a position to grant this set of rights. Arguably, through this provision of its terms of service, any Midjourney user who uploads a copyrighted work that they do not own or have a license to would likely be in breach of the terms. 

  1. Does copyright law permit your use? 

To determine whether copyright law permits the use you want to make, we highly recommend the framework of considerations and resources that you can consult by following the link above. In brief, these considerations are: (1) Is the work copyrightable subject matter?; (2) Is the work in the public domain?; (3) Is the work subject to a license that permits your use?; (4) Is your use a fair use? 

If none of those considerations provide a clear resolution to your question, seeking permission for the use you want to make is recommended.  

What resources do you recommend to stay up to date with current AI litigation? 

We’re actively watching copyright developments related to AI and doing our best to share information promptly. Our blog is a good place to look for regular updates. You can also follow us on social media: Bluesky and LinkedIn.

Professor Ed Lee’s Chat GPT is Eating the World is an excellent resource for updates as well. For example, you can find the latest maps of copyright lawsuits related to AI, both in the US and internationally.. Professor Lee routinely revises the map to reflect newly filed lawsuits, so be on the lookout for additional maps in the future.    

This case tracker from BakerHostetler also does a good job of providing status updates for ongoing litigation. Similar case trackers can be found through a straightforward online search.

For day-to-day updates to specific cases, there’s no substitute for watching the court dockets associated with those cases. We highly recommend CourtListener for this purpose — it gives you the ability to search for specific cases and explore and download relevant documents (e.g. complaints, answers, court orders, etc.).  

For example, here’s the docket for Bartz v. Anthropic, a case that’s being litigated in the Northern District of California. We’ll be on the lookout for additional ways to stay up to date with current developments in AI litigation and we’ll add them here whenever possible. If you know of an excellent resource that we should consider sharing here, please let us know at info@authorsalliance.org.

Scroll to Top