Skating between human magic and machine, the fragile fate of generative AI ‘works’

[Georgia Jenkins]

Skating between human magic and machine, the fragile fate of generative AI ‘works’

 Georgia Jenkins Friday, February 20, 2026 - AI, AI generated creations, Artificial Intelligence, author's own intellectual creation, copyright work, cultural production, Georgia Jenkins, German copyright law, human creativity, logos, originality

Photo by Ансплэш Степана on Unsplash After all that, here’s the only creature truly equipped to skate the fine line between human and machine

For those following the 2026 Winter Olympics ice dance-related drama, the magic of the French team’s free dance from the ‘The Whale’ and rhythm dance set to Madonna’s ‘Vogue’ was nothing short of spellbinding, earning France a gold medal. The controversial duo may have dominated headlines, but two Czech siblings had an equally powerful effect on the “magic” of human artistry in ice dance. For the rhythm section of their programme, the pair danced to ‘‘Thunderstruck’ AC/DC and a song generated by AI, One Two by AI (of 90s style Bon Jovi).

While the pair placed 17th, backlash erupted online. Outrage that echoes concern in copyright circles on the status of generative AI ‘works’. The German first instance courts (Kneschke v. LAION; GEMA v. OpenAI; Lehmann) have likewise not hesitated to weigh in on generative AI (IPKat here, here, and here). Naturally the Munich court was poised to respond to a complaint on whether copyright subsisted in three logos generated by AI through user prompting pursuant to s.2(2) of the German Copyright Act. While ice dance seemingly skates on the edge between human and machine, the German court met the same entanglement in copyright law with a far starker answer.

 

Human creativity as a precondition for generative AI ‘works’

The court affirmed that the concept of a work of authorship is clear from settled CJEU case law. It is an autonomous concept of Union law that must be interpreted and applied uniformly and comprises two joint elements: (1) originality; (2) a ‘work’. Originality requires the work to be the author’s own individual creation and to reflect their personality by expressing their free and creative choices. To the extent that technical constraints determine the work and leave no room for artistic freedom, it is not an original work.

Drawing from German copyright commentary, the court explained that generative AI output can meet this threshold if it evidences ‘human creative influence’ despite the ‘software-controlled process flow’. The court explained that this requires ‘human intervention’, meaning that some degree of human-creative influence is necessary alongside the selection of AI-generated responses. Additionally the court leaned on the analogy of 'AI as a tool' to distinguish between users selecting content from AI ‘proposals’ versus user prompts demonstrating free and creative choices.

When AI is used as a tool, that court reflected that this demonstrates the user’s ‘personal choice’ as opposed to the AI’s technical functions predetermining the output. This means that open-ended user prompts are insufficient to demonstrate originality in a work of authorship. Put simply, human creative choices need to dominate the output. According to the court, this reflects traditional justifications that copyright rewards and protects human creativity, not investment, time or diligence.

 

A spectrum of human input in AI-generated logos

The plaintiff argued that they used AI as a creative tool, akin to a sculptor, where creativity comprises an iterative artistic practice of reflection and revision. In response, the defendant’s argument centred upon the relationship between the user and the AI, explaining that the human being does not create, but acts a source of ideas. Further the defendant hinted that the minimal effort required to generate AI content supports denying it copyright protection.

Unsurprisingly none of the three logos were found to be original works despite a spectrum of prompting.

At one end, the ‘laptop book’ logo was generated by a 2-line description to create a ‘simple but unusual’ logo for a website where legal texts can be read. This was insufficient to discern the creative development of the author’s personality.

In the middle, the ‘envelope-before-columns’ logo involved around 1,700 characters that were ‘formulated and tested’. While this was a long and carefully crafted prompt, the court reiterated that time expenditure is not a requirement of originality. Focusing on the detail, they found that the descriptions were too general (e.g. ‘design an original, abstract logo’; ‘modern, minimal, distinctly original’; and ‘clean flat design with custom geometric abstraction”). In effect, many of the creative decisions were left to the AI (e.g. ‘waves, motion lines, ray’; ‘base colours […] and others if you deem them a good fit’). This was considered comparable to giving instructions to a human developer to create a logo.

Towards the other end, the handshake-and-bell logo was a little more specific, comprising iterative prompting, revision and amendments (e.g. ‘those fingers must be white skinned, please’; ‘the last image seems to be broken. Recreate it please’). However, the court still found these to be technical or minor changes, meaning that the logo was designed by the model and not shaped by the user’s personality.

 

Comment

The judgment avoids discussion on whether the logos themselves, if created without the assistance of generative AI, would be protected by copyright. It focuses on the relationship between user prompting and the work’s final form. The user prompting spectrum (almost certainly party-engineered) raises questions over the free and creative choices (prompt writing, selection and correction) that potentially comprise generative AI-assisted output.

First there is the hint of a technology-specific evidential threshold for users to demonstrate process-side input. This is in stark contrast to other authors, who rely on the form of the work with courts inferring the existence of free and creative choices. In Levola Hengelo, the CJEU emphasizes the form of the work (para. 40). Recently in Mio/Konektra the court explicitly requires national courts to ‘seek out and identify the creative choices […] to declare it protected by copyright as a work’. And yet the Munich court seems to apply a stricter standard for generative AI content. A stance that sits uneasily with Cofemel where the CJEU held that the same originality standard applies to all categories of works and that there are no additional criteria.

By extension, this Kat ponders the relationship between user prompting and seemingly ‘low’ levels of creativity. Painer confirms that even a simple portrait photograph, despite limited creative freedom, can meet the originality threshold. Further that when a work is constrained by functional requirements, there may still be room for genuine creative choice (Mio/Konektra). German copyright law clarified that originality, including low level creativity (Kleine Münze) applies equally to applied art following the German Federal Court’s Geburtstagzug judgement. Additionally where content lacks originality, s.72 German Copyright Act still carves out special protection for non-original photos. However, the Munich court, citing German copyright commentary, notes in obiter, that even ancillary copyright protection is excluded where a user merely selects AI ‘proposals’ without sufficient creative human intervention.

More broadly the ‘AI as tool’ rhetoric relies upon the construction of AI as a neutral instrument that enables human creativity. Yet cultural production theorists suggest that AI should be viewed through the relational lens of ‘distributed creative agency’ between humans and non-human actors. It means evaluating how AI reshapes creative practices and social structures, and in turn, revisiting an originality threshold already challenged by cumulative, collaborative and community-driven forms of creativity.

As one commenter explained, the generative AI ice dance song was ‘completely soulless… the most generic ‘rockesque’ thing you can think of’, and perhaps some AI-generated logos (and even certain ‘simple’ portraits) invite a similar response. So this Kat very cautiously recalls the old ‘labour, skill and judgment’ test in the UK as a potential starting point to consider how the copyright system recognises creativity-relevant work within data infrastructures. At least then it might get us closer to recognising the often invisible work that shapes these outputs even if that line of thought is, for now, very much on thin ice.

Do you want to reuse the IPKat content? Please refer to our 'Policies' section. If you have any queries or requests for permission, please get in touch with the IPKat team.