On 2 March 2026, the US Supreme Court declined to hear the appeal in the Thaler case: an image purely generated by AI is not protected by copyright. In the US, that is final. Yet at the same time, Chinese courts grant protection to AI images, and Europe is refining its own doctrine. Three blocs, three philosophies. Here is the verified comparison, then what it means in practice for a Belgian entrepreneur, down to your book cover made with DALL-E.
The case lasted four years. In 2022, the Copyright Office refuses to register "A Recent Entrance to Paradise", an image Stephen Thaler credited to his machine, with no human author. In March 2025, the appeals court confirms: US law requires a human author. On 2 March 2026, the Supreme Court declines to hear the appeal. The rule is settled.
An image generated without substantial human input (ChatGPT, Midjourney, Adobe Firefly) falls into the public domain in the US. You publish it, anyone can reuse, modify and sell it, legally and without paying you. An important, confirmed nuance: if you strongly guide the process (precise prompts, iterations, edits), partial protection can cover your human contribution. The boundary is assessed case by case.
The common thread, which already answers part of your questions: the AI is never the rights holder. Where the blocs differ is the threshold of human input required, and how readily a judge finds it met.
| Situation | United States | China | EU / Belgium |
|---|---|---|---|
| AI alone, no human input | No copyright (Thaler, final 2026) | No copyright | No protection |
| AI + substantial human input | Human parts protectable | Protectable if creative effort is proven | Protectable if "author's own intellectual creation" |
| Burden of proof | On the claimant | Strict: show the real process | Case by case (free, creative choices) |
| Can the AI be the author? | Never | Never | Never |
United States: a strict human-authorship doctrine, now locked in. China: more open. As early as 2023, the Beijing Internet Court (Li v. Liu) recognised copyright in a Stable Diffusion image, because the user made creative choices (design, selection and arrangement of prompts, settings, iterations). In 2025, judges raised the evidentiary bar: you must show the real creative process, not an after-the-fact reconstruction. EU and Belgium: an approach centred on human creativity. The EU Court of Justice requires an "author's own intellectual creation". A purely machine output is excluded. The European Parliament, in its 2026 work, confirms that content without a human author should remain ineligible, and proposes labelling; the EU AI Act already requires transparency on data and marking of AI content.
The Belgian principle follows the EU: protection arises from your human creative input, not from clicking "generate". Applied to your examples:
You ask the right fundamental question. Let us separate the settled from the proposal.
There is no "right of the AI". An AI is not a legal person: it can own nothing, sign nothing, be paid or taxed. The "electronic personhood" proposal raised in the European Parliament in 2017 was dropped after opposition from more than 150 experts. The only possible rights holder remains human (or the company that employs them).
Our reading, where the law is still moving. What you call a "right of conception" is not a new category to create: it is the recognition, case by case, of your human directing contribution, namely the elaborated prompt, the selection, the iteration and the editing. That is exactly what China already protects subject to proof, and the direction European analysis is taking. Our proposal for businesses: think in terms of a "design file". Keep a record of your creative choices (briefs, versions, retained prompts, edits) just as one keeps a designer's sketches. That file, not the tool, makes the difference before a judge or a client.
Take your case: you create your book cover with DALL-E, the book sells well. Can the AI claim money from you, as an illustrator would have?
No. The AI has no legal existence to claim anything. Nor does the provider: OpenAI's terms assign you the rights to the output and impose no royalty on your sales. You pay a subscription or usage, like a tool, not a co-author. The illustrator invoiced you a fee and paid their own taxes on it; the AI charges nothing on the book's success. On tax, it is clear-cut: the AI is not a taxpayer. It is you (or your company) who collect the book's revenue and are taxed on it, like any author or business income in Belgium. The AI subscription cost is a deductible expense.
The AI will not send you an invoice, but the absence of cost has a price. Your purely generated cover may be protected by no copyright: a competitor can reuse it. It may also accidentally reproduce protected third-party elements and expose you to a dispute. Finally, the provider's terms can change. The right reflex: add real, documented human input, and run a review before publishing or printing.
And tomorrow? The idea of a "tax on machine output" (the famous "robot tax") recurs in the debate, but none has been adopted to date. If it came, it would hit the company that uses the AI, never the model itself. The principle stays stable: humans produce, humans collect, humans are taxed.
In the US as in Belgium, an output with no human input is not protected. China can protect, but demands solid proof.
Your creative choices (prompts, versions, edits) are the raw material of any right. Keep them as a design file.
No royalties, no tax on "its" output. You collect and you are taxed, as before AI.
Human review, third-party-element checks, read the provider's terms. Linked to our guide on AI code ownership.
Molderez Consult SRL helps you organise the traceability of your human input, your contracts and your pre-publication checks, to protect your content and limit risk.
Frame my AI creationsTransparence : cet article a été rédigé avec l'aide de l'intelligence artificielle, puis relu par Molderez Consult SRL. Information générale vérifiée le 19 juin 2026, elle ne constitue pas un avis juridique ni fiscal individualisé ; les passages signalés comme « notre lecture » ou « proposition » sont des interprétations, le droit étant encore en évolution.