The post points to a Munich local court decision holding that purely AI-generated material is not protected by copyright under EU law. The key distinction is not “AI involved” versus “no AI involved.” It is whether the system acted like a tool under meaningful human creative control, or whether the model itself supplied the expressive choices. Several people noted this is consistent with older copyright logic around automated recordings, generative systems, and the basic Berne Convention idea that copyright attaches to original human authorship.
What actually got traction was the mess this creates for mixed human and AI workflows. Prompting alone does not seem to count. Selecting from multiple model outputs likely does not count either. Human edits may only protect the edits, not the untouched machine output. That leaves a fuzzy but important operational rule: if you want copyright, you need evidence that a human shaped the expressive result, not just the request. People expect that to matter most in code, design, and media businesses that assume they can both ship AI output and enforce licenses around it.
The thread also landed on a second point that is easy to miss. “No copyright in the output” does not mean “free of copyright problems.” An AI-generated translation of a copyrighted novel may lack its own fresh copyright, but it can still be constrained by the original novel’s copyright because it is a
derivative work. The same logic applies to code and media that too closely tracks existing protected material. So the near-term consequence is asymmetric. AI output may be hard to protect, yet still easy to infringe with.
A lot of the practical anxiety came from software. If code is largely machine-generated, copyright-based licenses like
GPL,
MIT,
BSD, and
Creative Commons may be unenforceable over those portions because there is no protected work to license in the first place. But that does not poison an entire project automatically. Human-authored structure, arrangement, surrounding code, and other original contributions can still be protected. The result is a
provenance problem more than a philosophy problem. Teams will need to know which parts were authored, which were merely assembled, and which are legally bare output.
The dominant read was that this is not a radical break so much as a preview of ugly edge cases. The decision is from a lower German court, not a final EU-wide settlement, and people expect years of litigation around how much human contribution is enough. Still, the direction of travel looks clear. Pure AI generation is a weak foundation for ownership claims. Businesses that need exclusivity will have to design workflows that leave a real human creative trail.
mathstodon.xyz