Artists who have begun to incorporate AI tools into their practice are encountering a question that did not exist in quite the same form a decade ago: who, if anyone, owns the result? The answer depends on where the work was made, how it was made, and how much of the final image, text, or sound can be traced to a human decision rather than a machine output. Readers of Ochre & Circuit have written to ask whether a piece generated with the help of diffusion models, large language models, or AI-assisted editing tools can be copyrighted, sold, or licensed with confidence. The short answer is that confidence is premature in most jurisdictions, though understanding the current state of the law allows artists to make more informed choices.

This article lays out what is reasonably settled, what remains contested, and what practical steps an artist can take while the legal frameworks continue to develop. It is written for working artists rather than lawyers, and it does not substitute for advice from a qualified intellectual property attorney, particularly for anyone preparing to sell, license, or exhibit AI-assisted work commercially.

Why Copyright Rules Are Still Unsettled for AI-Assisted Work

Copyright law in most countries was built around a simple premise: a human being creates a work, and that human (or their employer, under certain contracts) owns the rights to it. AI-assisted creation disrupts that premise because the degree of human involvement varies enormously from one project to the next. A photographer who uses an AI upscaling tool on a photograph they took themselves is in a very different legal position than someone who types a single prompt into a text-to-image model and publishes the first result.

Legislators and courts have not caught up with the pace of the technology. The United States Copyright Office has issued guidance and individual rulings since 2022, but it has done so through case-by-case review rather than comprehensive legislation. The European Union's AI Act, which entered into force in stages beginning in 2024, addresses transparency and risk categories for AI systems but does not resolve authorship questions for creative output. This gap between technological capability and legal clarity means that an artist publishing AI-assisted work today is operating in a space where the rules may be reinterpreted before the work has even finished its commercial life.

Several practical consequences follow from this unsettled state. Registration decisions by copyright offices can be appealed and reversed. Licensing agreements drafted today may need revision if a governing law changes. Platforms that host or sell art, including print-on-demand services and stock libraries, have each adopted their own internal policies about AI-assisted submissions, and those policies shift as the companies respond to legal developments and user pressure.

Human Authorship as the Current Legal Anchor

Despite the uncertainty, one principle has held steady across most jurisdictions examined so far: copyright protection generally requires a human author. This was made explicit in the United States in the case Thaler v. Perlmutter, in which a work generated entirely by an AI system without human creative input was denied registration. The courts and the Copyright Office treated the absence of human authorship as disqualifying, regardless of the complexity or originality of the output.

A more instructive case for working artists is Zarya of the Dawn, a graphic novel in which the author used Midjourney to generate individual images but arranged, selected, edited, and combined them into a narrative sequence. The US Copyright Office ruled in 2023 that the individual AI-generated images could not be copyrighted on their own, but the selection, arrangement, and accompanying text created by the human author were protectable. This distinction between the raw output of a generative tool and the human-authored compilation or modification of that output has become a reference point cited in subsequent guidance.

Practically, this means the following factors tend to matter when assessing whether a human contribution rises to the level of authorship:

• Iterative prompting followed by selection among many outputs, rather than acceptance of a single first result
• Manual editing, compositing, or painting over AI-generated elements
• Combining multiple AI outputs with original human-made elements into a larger composition
• Written or recorded documentation of creative decisions made during the process

No current guidance specifies a precise threshold, such as a percentage of manual revision, that guarantees protection. Offices and courts are evaluating the degree of human creative control on a case-by-case basis, which is one reason documentation, discussed later in this article, has become so important.

How Jurisdictions Differ in Practice

Copyright is territorial, meaning that protection and registration rules differ by country, and an artist selling work internationally may need to consider several legal systems at once. The table below summarizes publicly stated positions as of this writing, though artists should verify current status with a specialist before relying on any of it for a commercial decision.

The UK's Section 9(3) is worth particular attention because it is one of the few statutory provisions written specifically to address works with no identifiable human author, attributing rights instead to the person who undertook the arrangements necessary for the work's creation. Whether this decades-old provision applies neatly to a modern diffusion model is a question legal commentators continue to debate, and it has not yet been tested extensively in court against contemporary tools.

Artists working across borders, such as those selling prints through an international platform or licensing illustration work to clients in multiple countries, should treat jurisdictional variation as a genuine business risk rather than a technicality. A work that receives registration in one country may face a different outcome if challenged in another.

Documenting Your Creative Process for Future Reference

Given that human creative control is the central question in most current rulings, keeping a record of that control is one of the more useful habits an artist can adopt. This is not a legal requirement in itself, but it creates evidence that may matter if a work's authorship is ever questioned, whether by a copyright office, a licensing partner, or a court.

A practical documentation habit includes the following elements:

1. Save intermediate versions of a piece, including early drafts, rejected AI outputs, and the sequence of edits made afterward
2. Keep a written log of prompts used, along with notes on why certain outputs were selected over others
3. Record the specific tools and model versions involved, since policies and capabilities change between software updates
4. Note any manual work performed afterward, such as repainting, color correction, compositing, or the addition of original elements
5. Retain timestamps or file metadata where possible, since dated records carry more weight than recollection alone

Some artists keep this record in a simple project folder with dated subfolders; others use screen recording software to capture an entire working session. Neither approach is officially mandated, but both produce a paper trail that a specialist can review if a dispute arises. An illustrator who later needs to demonstrate that a commercial client's image was substantially hand-edited, rather than a direct AI output, will find such records far more persuasive than a verbal account produced after the fact.

Questions Worth Asking a Specialist Before Publishing or Selling

Because the legal landscape shifts by jurisdiction and by case, a brief consultation with an intellectual property attorney before a significant sale, licensing deal, or exhibition can save considerable difficulty later. Artists preparing for such a conversation may wish to raise the following questions:

• Does the amount of manual editing performed on this piece meet the threshold this jurisdiction currently applies for human authorship?
• Do the terms of service for the AI tool used grant the platform any ongoing rights to the output, and do those terms conflict with an exclusive licensing agreement?
• If the work will be sold internationally, which jurisdiction's copyright rules will govern a potential dispute?
• Does the gallery, publisher, or platform involved have a written policy on AI-assisted submissions, and does the work comply with it?
• Should the documentation of the creative process be formalized, such as through notarization or deposit with a registration office, given the value of the work in question?

A specialist will also be able to advise on training data concerns, since some AI tools have been the subject of litigation over the data used to train their underlying models. This is a separate legal question from authorship of the output, but it can affect an artist's exposure if a client or buyer later raises concerns about provenance.

Common Mistakes

Artists new to this area tend to repeat a small number of avoidable errors. Treating a single AI output as finished work without any documented revision leaves little basis for a later authorship claim. Assuming that registration granted in one country will be honored identically elsewhere ignores the territorial nature of copyright. Relying on a platform's terms of service summary rather than reading the actual licensing language can lead to unpleasant surprises regarding who else may use or claim rights to the same output. Finally, waiting until a dispute arises to begin documenting a creative process is far less useful than building the habit from the first draft.

A Brief Summary of the Present Legal Landscape

At the time of this writing, human creative authorship remains the central requirement across the jurisdictions most relevant to working artists, and purely machine-generated output without meaningful human selection, arrangement, or editing is generally not protectable on its own. The United States, the United Kingdom, the European Union, and China have each taken slightly different approaches, and individual court decisions continue to refine how much human involvement is sufficient. None of this is final, and an artist should expect further rulings, legislative updates, and platform policy changes in the coming years.

For now, the most practical steps available are modest but real: document the creative process thoroughly, understand that registration outcomes vary by country, read the terms of service for any tool used in a commercial project, and consult an intellectual property attorney before a significant sale or licensing agreement involving AI-assisted work. These habits will not resolve the underlying legal uncertainty, but they will put an artist in a far stronger position whenever that uncertainty is eventually settled.