The panic that followed Anthropic's decision to add invisible watermarks to Claude's output under the EU AI Act transparency rules has revived an uncomfortable question for anyone publishing with generative tools: who actually owns machine-generated text, images and logos? According to one of the world's most-cited copyright scholars, the answer in the European Union is often nobody at all.
In an interview published by EUobserver on August 20, 2026, Vanderbilt Law professor Daniel J. Gervais walked through the consequences of EU copyright law's strictly human-centric foundation. His conclusion is bracing for a content industry that has spent three years mass-producing AI-assisted material: content that is entirely generated by artificial intelligence is not protected by copyright, and in many cases there is no right that anyone — author, publisher or employer — could claim or transfer.
The Verdict From Munich
Courts are already putting that principle into practice. In a ruling issued on February 13, 2026 (case 142 C 9786/25), the Munich Local Court (AG Muenchen) decided that three logos created with a generative AI tool enjoy no copyright protection whatsoever — because they lack any recognizable human creative imprint.
The case was unusually concrete. The plaintiff had used a generative AI chatbot to produce three logos, including a handshake between two people of different skin colors with a bell, an envelope in front of a columned building, and a laptop with a floating book marked with legal paragraph symbols. He worked with detailed, partly iterative prompts, published the logos on his website — and then sued an acquaintance who copied the graphics for his own site, demanding cease-and-desist and deletion under Germany's Copyright Act.
The court dismissed the claims. Even detailed or iterative prompts, the judges held, are not sufficient to make a "personal intellectual creation" objectively recognizable in the output. Relying on the Court of Justice of the European Union's harmonized definition of a "work," the court stressed that copyright requires originality reflecting the author's personality through free creative choices. Where the result is determined by technical processes rather than those choices, there is no protectable originality — and no exclusive rights to enforce.
A Provenance Mark, Not a Copyright
Gervais, who published an influential 2019 paper arguing that works lacking human creative choices belong in the public domain, told EUobserver that the same logic now extends across the AI ecosystem — with consequences most users have not considered.
Take the simplest case: an employee asks Claude to write a LinkedIn post and publishes it verbatim. "Nobody owns it," Gervais said. Feed an interview recording, research and background papers into a model and have it draft an article in your tone of voice, and the result is the same — while the recording itself may carry copyright in the spoken words, meaning publishing the AI-drafted piece could even infringe the speaker's rights.
And what happens when that article is handed to a publisher? Nothing transfers, because there is nothing to transfer. "When you put your name on an article that's written by ChatGPT or Claude, you're basically putting a provenance mark on it saying: I take responsibility for this," Gervais said. "I haven't written it, but I'm putting my name on it. That doesn't give you copyright, but it does give you liability for the content."
That asymmetry — liability without ownership — is the detail legal teams are only beginning to grapple with as AI-generated material saturates professional platforms. LinkedIn itself recently added an option to flag content as AI-generated after research estimated that more than a third of copy posted to the platform was produced by AI.
Why the Watermark Fight Missed the Point
The interview landed in the middle of a noisy controversy. Anthropic began adding invisible watermarks to Claude's text output to comply with the EU AI Act's transparency obligations, prompting user protests, account cancellations and a wave of workaround attempts. Some users discovered that copy-pasting into plain text does not strip the mark; others demonstrated that the watermark is tied to the model's next-word prediction and would require full rewrites to remove.
The debate over detection, however, largely overshadowed the ownership question — which, as Gervais notes, could quietly destroy the value of vast amounts of AI-produced content. If purely machine-generated work sits in the public domain, competitors in the EU can freely copy it, and the only defensible asset is the human contribution layered on top.
A summary of the interview circulated widely among developers this week, drawing hundreds of upvotes on Hacker News under the headline "Copyright does not protect AI-generated content in EU."
What It Means for Businesses
For companies, the emerging picture is double-edged. On one hand, logos, marketing copy and product text generated purely by AI can essentially not be protected on an exclusive basis in the EU — anyone may reuse them. On the other hand, the Munich ruling and the scholarly consensus point to a clear path to protection: demonstrable human creative decisions, made and documented before, during and after generation.
The practical takeaway for publishers and brands is to treat AI output as raw material rather than finished work. Editing, arranging and creatively reshaping machine drafts preserves at least a claim to authorship of the human layer. Publishing prompts and post-publishing a name alone secures nothing but responsibility.
As the EU AI Act's transparency provisions roll out across the industry, the gap between what companies think they own and what the law recognizes is set to become one of the defining legal battles of the AI era. For continuous coverage of AI policy as it happens, bookmark AI Buzz Wire, our daily hub for breaking AI news.
Stay ahead of the curve on AI regulation and its consequences — follow the breaking AI news and expert analysis at AI Buzz Wire.
Read more AI news →
