Are AI outputs copyrightable? The US Copyright Office view

In 2023, the US Copyright Office announced a plan to assess how copyright and artificial intelligence (AI) intersect. The Office consulted extensively with domestic and international stakeholders, interested parties and members of the public. The result of these consultations will be included in a report, published in three parts:

• ‘Part 1’ was published in July 2024 and discusses the issue of digital replicas, using AI to create replicas of people or their characteristics.

• ‘Part 2’, the subject of this post, was published in January 2025 and covers the question of whether copyright can be attached to works created by GenAI.

• ‘Part 3’ is expected before the end of Q1 2025 and is keenly awaited because it will deal with AI models using copyrighted work in their training data; a hot topic also being looked at by courts around the world.

This is the trillion-dollar question, and ‘Part 2’ of the report goes some way to answering it. ‘Part 2’ makes for interesting reading, and it’s clear that the Office carefully considered existing law, the submissions made to it during the consultation period and the workings of GenAI itself. The key questions we’ve extracted (and answered) after assessing the report are:

• Do we need “new law” that specifically covers AI copyright?

• What’s the relationship between copyright and how AI tools and materials are used?

• How are AI prompts viewed under copyright law?

Let’s get started and look at the answers.

First, the Office considered whether existing copyright law, contained in Title 17 of the United States Code has the framework and language necessary to assess whether copyright can be claimed in work created by GenAI. This is what was concluded:

This means no new intellectual property rights need to be created to deal with AI-generated work and no changes have to be made to existing copyright law; established copyright law can be applied to the current questions being presented by GenAI.

Note: The law on copyright goes back to 1886, when the Convention for the Protection of Literary and Artistic Works was signed in Berne (the Berne Convention)!

The Office discussed how GenAI tools are used, and whether the nature or extent of the use has an effect on whether copyright exists. It also looked at work which contains elements of both human-created and AI-generated material. These were the conclusions:

Using a grammar correction tool or AI thesaurus on your work won’t affect your copyright in the end result. The distinction between “assistance” and “creation” doesn’t depend on the AI tool used, rather it depends on how that tool is used.

Even if your work contains some AI-generated material, the portion of it that you originally created remains protected by copyright.

Discussing the central understanding that a human being must control the expression, rather than the idea, of a work in order to obtain copyright, the Office didn't set out rigid rules to determine what level of “control” was enough. However, the Office did give a solid and far-reaching view about AI prompts. Here's what was concluded:

This is a fundamental tenet, from which much of the Office’s logic flows. A human being must have control not only over what is created, but how it is expressed.

This will look at whether the contribution represents a mere “instruction” or whether it is an expressive work in its own right. There are no hard and fast rules saying what level or type of human contribution is required. Except that…

The Office determined that prompts essentially function as instructions that convey unprotectable ideas, much like in a recipe. While highly detailed prompts could contain the necessary expressive elements and originality to justify copyright protection of the prompts themselves, this does not mean that the result of the prompt is also covered by copyright. This is a vital point made by the Office and the report discusses this reasoning in depth.

On page 19, the Office says:

“The 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, prompts may reflect a user’s mental conception or idea, but they do not control the way that idea is expressed.” (*Emphasis added).

And on page 18, the Office states:

“The provision of detailed directions, without influence over how those directions are executed, is insufficient”…[to demonstrate that the human had sufficient control over the output].

The opinions in the report are based on current technology, and the Office admitted that the position may change in the future. A shift in position would be given particular consideration if the workings of AI systems become more transparent, or if the tools evolve to give users the overriding control over all of the expressive elements of a work. Below, we've highlighted an example of how this could play out:

• 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. For instance, if you adjusted a photograph that you took by asking an AI tool to change a colour or remove a blemish, or even wholesale convert it to a different style, so long as your original work is recognisable in the output, you retain copyright.

The report is directly relevant in the US, and its reasoning will likely be persuasive to other countries as they establish their own standards around AI-generated work.

Other countries can, and probably will, take different views on the copyrightability of AI-generated works and of the prompts used to create them.

Already, the UK recognises copyright in computer-generated works and an expert committee in Ireland has recommended that “new law” is required to deal with AI-generated work.

What does this mean? We may see a highly fragmented copyright regime as it applies to AI-generated material, as nations seek and introduce new legislation that specifically covers it.