When a person uses AI to generate an image, who created the image — the person or the machine? This question is at the heart of the AI-generated image copyright issues that are becoming increasingly urgent as more people use popular AI tools like Midjourney, DALL-E, and Google Gemini. Where do you draw the line between AI-created and human-created images and text? That hasn’t yet been clearly resolved. Expect a lot of legal activity in the near future as the U.S. Copyright Office and the courts try to sort it all out.
Current Copyright Stance on AI-Only Works
According to guidance from the U.S. Copyright Office published in March 2023, it is “well-established” that copyright protection is only available for content produced by human creativity. The term “author,” as used in copyright and Constitutional law, excludes non-humans. For example, there have been cases where selfies taken by a monkey or words said to be written by non-human spiritual beings did not qualify for copyright protection because they lacked human authors.
The same is true for images created by a machine, even when a person provides the machine with a prompt. The Copyright Office guidance says that when AI generates an image (or other type of complex work) from a human prompt, the technology, not the human, is executing the “traditional elements of authorship” – and, therefore, the image is not the product of human authorship and cannot be copyrighted. These AI-generated images are in the public domain.
However, if there is “sufficient” human involvement in creating an image that contains AI-generated material, then part or all of the work may be copyrighted. But how much is “sufficient”? That key question in AI image copyright law is still being worked out.
Human Input and Copyright: Where Does Creativity Start?
The Copyright Office says the question of how much human input is required to make an image eligible for copyright protection should be decided on a case-by-case basis. At the core of each decision is the question of whether the image was essentially created by a person with AI merely assisting, or whether the AI determined how the human’s instructions were to be carried out.
The potential exists to copyright an image if a user significantly modifies or customizes an AI-generated image. Copyright might be granted for either the whole work or just the human-created modification, depending on the extent of the human’s contribution to the work.
How to measure the contribution is still a gray area. One example: In 2022, an artist won an art competition with his AI-generated image “Théâtre d’Opéra Spatial. He claimed he made at least 624 revisions and prompts, and made adjustments to the image using Photoshop. He applied for copyright. After a hearing and two requests for reconsideration, the Copyright Office’s Review Board turned him down each time, concluding that the AI-generated image remained in substantial form in the final image even after all the additional prompts and modifications. The Board also said it did not have enough information to decide whether the artist’s adjustments would be copyrightable on their own, and that the artist should have disclaimed the AI-generated material in his copyright application, which he had refused to do.
Copyright Issues Around Training Data
While the question of the copyrightability of AI-generated images deals with the output of AI models, there are also copyright issues arising at the other end, from the input. AI image generators get trained on massive amounts of data, much of it copyrighted material. Some individuals and organizations are going to court to try to stop AI companies from using their images and words. In 2023, the New York Times sued OpenAI and Microsoft and Getty Images sued Stable Diffusion. In November 2024, five of Canada’s largest news companies filed a joint suit against OpenAI. The AI companies are claiming fair use as a defense.
Practical Takeaways for Using AI-Generated Content
Images that you create using AI image generators will most likely end up in the public domain, in which case, you won’t be able to protect them from being used by someone else. If you think your work is in the gray area of human/machine input, stay updated on copyright developments in AI-generated art ownership, check the terms of service for the image generators you are using, and be careful when using AI images in commercial contexts. Know that if you do apply for copyright protection for works containing both human and AI-created material, the Copyright Office’s policy is that applicants have a duty to disclose the inclusion of AI-generated material and to briefly explain the human contribution to the work.
Conclusion: A Legal Landscape in Flux
The Copyright Office is focused on bringing clarity to AI-related copyright issues and policies. It’s been seeking public input and has just started issuing a multi-part report to analyze the issues.
In July 2024, the Office released the first part of its report, which is on digital replicas. These are AI-created or altered videos, images, and audio that can create convincing but fake depictions of people. Still to come (as of December 2024) are sections of the report that will cover the issues mentioned here – whether materials created in whole or part by AI can be copyrighted and the legal implications of using copyrighted material to train AI models. Other sections of the report will address licensing and potential liabilities.
Litigation is also working its way through the courts, with nothing definitive decided yet. It could take as long as a decade to reach a final resolution if the cases go all the way to the Supreme Court. The cases where news organizations are claiming copyright infringement of their materials could resolve much sooner, though, through settlements involving licensing agreements. The Associated Press and some other news outlets have already made those kinds of deals with AI companies.
Other countries are taking different approaches to AI image copyright law. The UK, for one, provides copyright protection to computer-generated works. If and how these differences will be reconciled in the future remains to be seen.
In the meantime, AI is increasingly becoming a part of people’s everyday lives, with the ability to generate images and text just a click away on everyone’s screens. With the massive amount of AI-generated work out in the world, there needs to be clearer policies on copyright protection. Expect more guidance to be coming soon.