The AI Copyright Conundrum

Imagine spending hours crafting the perfect prompt to generate an AI masterpiece, only to discover someone else selling copies of it online. Who actually owns the rights to AI-generated work? A new report from the US Copyright Office attempts to answer this question, but the solutions may not satisfy everyone.

Last week the US Copyright Office released a 41-page report examining how current copyright laws address original works created with AI assistance. As someone who uses these tools regularly in my creative process, and who has a partner who uses AI to generate music, I was eager to understand what rights, if any, we AI-enabled creators might have. The answers, not surprisingly, are not so simple.

As it turns out, the authors of the report essentially determined that the current laws do not require updates. Apparently the existing rights and exclusions sufficiently protect the rights of (human) creators, and those rights at this time do not extend to the outputs of AI tools except in specific cases. But what does that mean exactly?

Understanding AI and Copyright: A Legal Framework

The intersection of artificial intelligence and copyright law presents interesting and complex challenges that stem from a fundamental requirement: human authorship. In the United States, copyright protection is rooted in the Constitution's Copyright Clause, which has consistently been interpreted to require human creators. Most other countries have historically attributed copyright only to humans as well, and it’s only recently that this distinction has required further clarification.

The Role of AI in Creative Works

While AI can serve as a powerful creative tool, potentially benefitting artists and creators of all levels and abilities, the legal framework in the US distinguishes between AI assistance and AI authorship and this is where the issue of ownership comes into play. When AI functions as an assistive tool, for tasks like colour correction or text refinement for example, it doesn't affect copyrightability. However, when it comes to generative AI outputs, whether it’s an image or a poem, the copyrightability depends greatly on the level of human control over the creative process. Simple prompts or mere selection of AI-generated content is generally considered insufficient to establish authorship, and courts have ruled that such minimal involvement doesn't meet the threshold for copyright protection.

(A Few) Pathways to Protection

There are nevertheless a couple ways that AI-assisted works may be copyrighted under the current framework. The first is when human-created inputs remain clearly perceivable in the final output. The other way is if someone creatively selects, edits, and arranges AI-generated material into an original composition. In both of these scenarios copyright protections may apply, though the individual AI-generated elements remain unprotected.

So in the first case, if I were to come up with an original melody, and I recorded it and uploaded it into Udio to use as the basis of a new song, and that melody was clearly recognizable in the final output, I would retain the copyright. In this example AI has enhanced my creative input.

Similarly, if I ask ChatGPT to write me a poem, and then I change the poem by extending it, rewriting parts, or otherwise editing it sufficiently so that the final poem has been creatively altered, then I would be able to claim copyright over the poem as a whole. However, I would not be able to claim copyright over the parts that were left unaltered!

As another example, if I took an image generated with Midjourney and used another program to change the background, or if I used it as part of a collage made up of other (potentially AI-generated) images, I would be able to copyright the final artwork - but I would not be able to copyright the parts that were generated in Midjourney (or by any other generative AI tools).

Global Perspectives

The US approach to copyright aligns with current international consensus, as most major jurisdictions, including South Korea, Japan, the EU, and China, maintain that copyright requires human authorship. The United Kingdom stands as a notable exception with its existing provision for AI-generated works, though its practical application remains unclear and is currently under review, while nations like Canada and Australia are also reevaluating their approach.

In the UK, copyright protection has for a long time been granted for Computer Generated Works (CGWs) and it has it has generally been extended to AI, though this is currently under review. In the case of a general-purpose AI that generates an output in response to a user prompt, the person who inputted the prompt is considered the "author". The UK Government has expressed a preference to remove these rights however, at least for images and text, because the current legal definition of CGWs requires the work to be "original," which is generally defined as an "author's own intellectual creation”, which requires more than a simple prompt of say, “a poem”. In other words, the UK law as it currently stands is problematic if not paradoxical.

The Canadian government is currently conducting a similar review of its policy. Current Canadian copyright law suggests that authorship must be attributed to a “natural person who exercises skill and judgment in creating the work”, with this wording being reflective of the fact that the Canadian Copyright Act ties the term of protection to the life and death of an author. While a person might conceivably put “sufficient skill and judgement” into an AI-assisted project to claim authorship, it's more difficult to make that case if they’re putting simple prompts into tools like ChatGPT or DALL-E and getting the AI to generate the content.

The Case for Current Law

The arguments against changing existing US copyright law insist that the current framework adequately accommodates AI-assisted creativity without requiring legislative amendments, based on the ages-old premise that to be copyrighted, it must be original, and to be original, it must be created by a human author. On top of that, critics of AI-generated content argue that additional protections are unnecessary given that AI development is thriving despite AI-generated content not receiving copyright protection. They further warn that extending copyright to AI-generated content could potentially dilute human creativity and reduce incentives for human authors. As a result of these arguments, the concept of special protections for AI-generated content has been dismissed as potentially harmful to human creators, for now.

The Big Unanswered Question

The fundamental tension in this debate lies in an apparent legal paradox: if copyright isn't granted to the person who prompts the AI to generate the output, then who owns it? Most AI companies' terms of service explicitly state that users own both their inputs and outputs, meaning they don't claim ownership of the generated content (even if they claim a worldwide perpetual license to reproduce it). However, if these terms won't hold up in court, creators are left in a precarious position.

The music industry might offer a relevant parallel. When sampling technology emerged, it created similar questions about ownership and creativity. The industry eventually developed a complex but functional system of licenses and rights management. Could a similar framework evolve for AI-generated content? Perhaps, though that would require first resolving disputes (and court cases) launched by artists whose copyrighted works were used to train these AI models in the first place.

For now, creators working with AI tools face uncertainty. While the Copyright Office's report provides clarity on current law, it neglects an urgent need for solutions that address the commercial concerns of professional AI artists, as well as those of designers and creative agencies currently using these tools to produce client work. As with so much else in this current era of technological disruption, there are still many loose ends to be tied.

Research and editing assistance for this article was provided by the following generative AI tools: NotebookLM, Perplexity, ChatGPT (o3-mini), and Claude. Cover art was created with Midjourney.