U.S. Copyright Office Report on AI and Copyright

Open Source AI
1

Yesterday the U.S. Copyright Office has just released a report on the copyrightability of works created using generative AI. Some key points:

  1. Existing copyright law is sufficient to address AI-related copyright questions—no new legislation is needed.
  2. AI assisting human creativity (rather than replacing it) does not prevent copyright protection for the resulting work.
  3. Human authorship remains essential—even if a work includes AI-generated elements, copyright applies only to human contributions.
  4. AI-generated content alone cannot be copyrighted unless a human has exerted sufficient creative control over its expressive elements.
  5. Determining whether a human’s contribution is sufficient for copyright must be assessed on a case-by-case basis.
  6. Prompts alone are not enough to establish human authorship—they serve as instructions, not expressions of creative control.
  7. A human can claim copyright over creative modifications, curation, or arrangement of AI outputs.
  8. The Copyright Office does not support additional or special protections for AI-generated content.

Full report: https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf

2

It appears that this new report reaffirms the standards regarding the copyrightability of AI-generated works that the Copyright Office set out in its guidance issued two years ago. Because the Office has concluded that no new legislation is necessary, we can likely assume that the criteria for copyrightability of AI-generated works in the United States are now almost settled.

Interestingly, there was a lawsuit in China last year concerning the copyrightability of images generated with Stable Diffusion, and the court recognized copyright in images created simply by entering prompts.

According to the U.S. Copyright Office’s position, no matter how complex the prompt input might be, humans cannot exercise creative control over the final output; the traditional creative and expressive elements are determined by the AI rather than by humans. Therefore, such works are not considered the product of human creativity and are thus not copyrightable. In contrast, in the Chinese lawsuit, the court held that through the human’s complex prompt input and other actions, the human’s selection, arrangement, and individual judgment were embodied in the AI-generated work. Thus, the court deemed the AI-generated work to be a human creation and therefore copyrightable. (Note that in Japan there is still no case law on this matter. However, based on the Agency for Cultural Affairs’ guidance, it appears that if the prompts contain sufficient human creativity, or if there is creativity and ingenuity in repeated attempts or in combining multiple AI-generated outputs, that might be enough to satisfy the requirement. This could be seen as an approach somewhere between the positions of the U.S. and China.)

When discussing Open Source AI, the divergence in how different jurisdictions interpret the copyrightability of AI-generated works likely comes down to how synthetic data is treated, don’t you think?

In the United States, I believe there is very little scope for copyright to arise in synthetic data. However, in China, it seems there may be a possibility that the same process used to create synthetic data could result in copyright being recognized.

2 Likes