Generative AI unsettles art's two oldest assumptions: that a work has an identifiable human author, and that the material an artist creates belongs to them unless they say otherwise. The disputes that follow cluster around two questions. First, is it acceptable for AI developers to train their systems on vast datasets of existing images without the originating artists' consent or compensation? Second, when a machine produces an image, who—if anyone—is the author, and who answers for it? No court or creative institution has settled these questions, and several are explicit that they cannot yet be answered definitively. What can be done is a disciplined way of asking them, which is what this article provides.

The Core Conflict: AI Training Data and Artist Rights

The point of friction is structural rather than incidental. Generative image systems learn patterns from very large collections of existing visual work, and much of that material was made by artists who never agreed to this use. From the artists' side, this looks like unauthorized appropriation: work created under copyright, scraped into datasets, and used to power a commercial product, with no consent, no compensation, and often no notification at all. A paper in Open Praxis on generative AI, authorship and ownership captures why the problem cuts across domains, noting that it raises "legal and ethical challenges of authorship and ownership, especially in relation to copyright laws and AI-generated content."

A parallel worry concerns style. Artists argue these systems don't merely store or copy images—they internalize the distinctive visual signatures that make an artist's work recognizable, allowing the model to produce imitations of an individual's creative identity on demand. That is what makes the conflict feel existential rather than merely transactional: what is borrowed is not a single picture but a lifetime of artistic development. How developers answer these claims differs by jurisdiction and case, and legal interpretations are still evolving, so this account reports the artist-side framing without declaring it the settled legal position. What is already clear is that transparency—knowing what a model was trained on—is the precondition for any honest assessment.

Copyright and Beyond 2024– Generative AI and the Law for Artists and AuthorsTo provide a direct, contextualized example of the legal and ethical discussions surrounding AI and copyright for artists.Watch on YouTube

Authorship in the Age of AI: Who Gets the Credit?

Traditional authorship was never just about who made something; it bundled together credit, responsibility and ownership. The International Committee of Medical Journal Editors states the principle plainly: "Authorship confers credit and has important academic, social, and financial implications. Authorship also implies responsibility and accountability for published work." The two halves travel together. Whoever takes the credit must also answer for the work—its accuracy, its integrity, its consequences.

Generative AI strains that bundle. When an image emerges from a model's statistical processes, the chain of substantive human intellectual contribution can be thin, and the chain of accountability thinner still. Is the person who typed a prompt an author, a client commissioning a work, or something else entirely? Writing in the animation studies community, a scholar on generative AI's creative frontiers offers one framing: whether a generative AI is used to generate a film, "the 'prompter' turns into more of an art director rather than an animator." The metaphor is illuminating because it concedes the prompter does contribute—art direction is real creative labor—while acknowledging that the labor's nature has changed, and much of the execution lies with a system no one can hold responsible in the way an author is held responsible.

That accountability gap is the crux. A human author can be asked to defend choices, correct errors, and bear professional consequences. A model cannot. This is why, as the Open Praxis paper argues, the field needs "new mechanisms to identify generative AI use"—the traditional signals of authorship no longer do the work on their own.