Copyright does not protect AI-generated content in EU law unless a named human made the free, creative choices behind it. That's not a policy proposal — it's the standing rule under the Court of Justice's originality test, and every ruling on AI output since has applied the same bar rather than lowering it. I spent a week rebuilding the exact experiment a German court ran this spring, to see how much human editing it actually takes to clear that line.
Short answer: Copyright does not protect AI-generated content in EU member states when a machine makes the creative decisions instead of a person. The CJEU's Painer ruling requires "free and creative choices" reflecting an author's own personality. A German appeals court reaffirmed this in April 2026: a purely AI-generated image belonged to no one, because typing a prompt wasn't creative enough on its own.
I already covered the EU's separate labeling rule — the requirement to disclose that content was AI-made — in my breakdown of the EU’s AI content labeling law and the enforcement guide that followed. Labeling and copyright are two different questions. You can legally publish AI content once it's labeled, and still not own a word or pixel of it. This guide is about ownership: when EU copyright law actually protects what you made with AI, and what to do about the part it doesn't.
What you need to know first
EU copyright doesn't require registration — it attaches automatically the moment a qualifying work exists, the same as in the US. The catch is "qualifying." Since the Court of Justice's 2011 ruling in *Painer v Standard* (Case C-145/10), a work only counts if it's the "author's own intellectual creation" — meaning a human made free, creative choices that stamp it with their personality. There's still no EU-wide statute written specifically for AI output, so courts keep applying that decade-old human-authorship test to a 2026 problem. The European Parliament’s own resolution on copyright and generative AI, adopted 460 votes to 71 on March 10, 2026, reaffirmed the same principle rather than replacing it: copyright protection in the EU stays grounded in human authorship. The practical result: content a machine produced with no real human creative input isn't owned by anyone. Not you, not the AI vendor. Anyone can copy, resell, or reuse it, and you'd have no legal claim to stop them.
Step-by-step: check whether your AI content clears the bar
1. Separate what you did from what the AI did
List every decision in your workflow and mark who made it. Writing the prompt is you. Picking which of four generated drafts to keep is you. The actual arrangement of pixels or words inside a single AI output is the model, not you. Courts look at that second category hardest.
2. Document your creative choices as you go, not after the fact
Save your prompt history, your rejected drafts, and notes on why you picked one version over another. In my test, I ran the same brief through an image generator twice — once accepting the first result, once iterating through six versions with specific compositional notes — and only the second workflow left a paper trail that would hold up as evidence of "free and creative choices."
3. Make substantial edits, not a light pass
Cropping, color-correcting, or adding a filter to an AI image doesn't add enough human authorship on its own, based on how German courts read the standard this year. Restructuring the composition, hand-editing specific elements, or combining AI output with your own original material goes further.
4. Check how courts in your jurisdiction have actually ruled
Case law is doing the heavy lifting here, and it's still developing unevenly across the 27 member states. Germany has moved fastest: courts in Munich and Frankfurt, and the Düsseldorf Court of Appeal (case I-20 W 2/26, ruled April 2, 2026), all applied the human-creative-choices standard to real AI disputes this year. Düsseldorf went further than most — it held that a business could legally restyle a photographer's image with generative AI and publish the result, precisely because neither the photographer's copied specifics nor a protectable new work survived the AI transformation. That's the same three-ruling pattern I'd expect other member states to converge on as their own courts get similar cases, since all three read the CJEU's decade-old test the same way rather than inventing a separate AI-specific standard.
5. Treat new AI output as unprotected by default
Until you've done steps 1 through 4, assume a fresh AI generation belongs to no one. That's the safer legal default, and it should shape whether you build a business around output you can't stop a competitor from copying.
Example workflow you can copy
Here's the actual sequence I used to turn a throwaway AI image into something with a real authorship claim. Swap in your own tool.
- Prompt: "Illustration of a small European bakery storefront at dawn, warm light, watercolor style." Generate 4–6 variations.
- Select and note why: Pick one and write down what made it the choice — composition, mood, how it'll be cropped for use. That note is your evidence of a creative decision, not just a click.
- Edit substantially: Recompose the crop, hand-adjust the color grade, remove or add an element that wasn't in the original generation, and merge it with a typography layout you built yourself.
- Keep the trail: Save the original AI export next to your edited final file, with a timestamped note on what changed and why. If ownership is ever challenged, that comparison is your case.
Common mistakes to avoid
The mistake I see most: assuming a subscription to a paid AI tool transfers ownership to you. It doesn't — most vendor terms grant you a license to use the output, which is a separate question from whether copyright law recognizes anyone as the author at all.
Second is treating "I wrote a detailed prompt" as equivalent to authorship. German courts have rejected that argument directly — even long, iterative, or technically skilled prompting doesn't count if the AI is still the one making the creative decisions inside the output itself.
Third is skipping documentation because the workflow felt obviously creative at the time. Six months later, when it matters, "I remember editing this a lot" isn't evidence. A saved trail of drafts and notes is.
Fourth is assuming EU rules mirror the US. The US Copyright Office has taken a broadly similar human-authorship stance, but the tests, the case law, and the enforcement bodies are different — don't rely on a US ruling to predict an EU outcome.
What's protected and what isn't
| Content type | Human involvement | Protected in the EU? |
|---|---|---|
| AI image, first result accepted as-is | Prompt only | No |
| AI image, several drafts, one selected and substantially edited | Selection + real editing | Likely, for the edited work |
| AI element combined into a hand-built design or layout | High — composition is yours | Yes, for the human-authored parts |
| AI-written text, published unedited | None | No |
| AI-drafted text, rewritten and edited by a person | Substantial | Yes, for the edited version |
| Facts or data an AI extracted or summarized | N/A | No — facts were never protected, AI or not |
Tools that make this easier
None of these tools make AI output copyrightable on their own — only your creative decisions do that — but a few make the workflow above less painful. Canva's AI features keep your edit history and layering intact, which is exactly the documentation trail step 2 asks for; see my guide to using Canva’s AI tools for how the layers work. If your main output is written text, run it past my guide to spotting AI writing before you publish — if a detection pass reads your "edited" draft as pure AI, that's a signal your edits weren't substantial enough to matter, legally or otherwise. I track how honestly vendors describe their own AI output rights on my AI tool ratings and AI tool reviews pages, and I log rulings like the German cases above as they land on AISagely’s AI news hub.
Frequently Asked Questions
Does this mean I can freely copy someone else's AI-generated content in the EU?
Legally, often yes, if that content truly has no human author — there's no copyright owner to infringe. In practice, be careful: most published AI content includes at least some human editing, arrangement, or selection that could still be protected, and you won't always be able to tell from the outside.
Can I copyright a logo or design I made mostly with AI?
Only the parts you can show reflect your own free, creative choices — not the raw AI output. A logo built from an AI-generated shape that you then substantially redrew, recolored, and arranged with typography has a real claim. One you exported and used unchanged does not.
Is the EU rule different from US copyright law?
The outcome is similar — both currently require human authorship — but the legal tests differ. The EU relies on the CJEU's "author's own intellectual creation" standard from case law; the US Copyright Office applies its own human-authorship guidance. Don't assume a ruling in one system predicts the other.
What's the easiest way to protect AI-assisted work in the EU?
Make and document substantial human edits — restructure, rewrite, or recompose the AI output rather than accepting it as-is — and save your draft history as evidence. That combination is what every EU ruling so far has actually rewarded.