AI has opened legal questions that existing legal systems were simply not designed to answer. Copyright law was built around a human being who creates. When the creation is wholly or partly generated by a machine, the answers are not straightforward. Here is a map of the key questions — and what is known (and what is not) as of today.

The debate is not settled, and the answers depend on jurisdiction and the specific creative process involved.

Position in the EU: The EU Copyright Directive does not recognise copyright over works generated exclusively by AI without significant human creative contribution. That contribution must be real and demonstrable — not merely writing a prompt.

Position in the US: The US Copyright Office has repeatedly denied registration for AI-generated works without demonstrable human authorship. The case Thaler v. Vidal (2023) confirmed that AI systems cannot be authors in the legal sense. The Copyright Office has since issued guidance stating that AI-generated content — without human selection, arrangement, or creative expression — is not protectable.

The grey zone: When an artist uses AI as a tool within a creative process involving significant human contribution — selecting elements, editing, composing, extensive retouching — protection may exist over that human contribution. There is no clear rule about what “percentage” of human intervention is enough, and there likely never will be, because the criterion is qualitative, not quantitative.

Practical recommendation: If you want to protect work that includes AI elements, document your creative process: screenshots of intermediate stages, exports of manually edited versions, a record of compositional decisions. Documentation is your proof of creative contribution.

Training data and artists

AI models (Midjourney, Stable Diffusion, Suno, etc.) were trained on artists’ works, often without explicit permission and without compensation. This has generated significant lawsuits in the US and Europe:

  • Andersen et al. v. Stability AI (US, 2023–2025): illustrators against Stability AI, Midjourney and DeviantArt for unauthorised use of their works in training.
  • Getty Images v. Stability AI (UK and US): for using millions of watermarked Getty images without a licence.

The EU AI Act (in force 2024–2026) includes transparency obligations for AI providers on training data. High-risk models will have to publish summaries of the data used, which may make future claims easier to pursue.

As an artist, you have the right to opt out of training data in some services. Adobe Firefly is marketed as trained exclusively on licensed or public-domain content — a market argument that many artists weigh when choosing which tools to use.

Publishing and selling work made with AI

Three situations carry very different implications:

1. Work entirely generated by AI and published as your own. High legal risk, high reputational risk. It is not illegal in itself in most jurisdictions, but you hold no copyright over the work — and if it comes out that you presented AI output as entirely your own without disclosure, the damage to your professional reputation can be lasting.

2. Work with AI elements integrated into a creative process. The zone where most artists are working today. Legal in most contexts if disclosed when relevant — in open calls, exhibitions, distribution platforms. Galleries, publishers and streaming platforms may each have their own policies, which you need to check case by case.

3. Your own work used to train models without consent. If you discover your work has been used without permission, you have options: opt-out via platforms that allow it (some repositories and social networks have added this), direct claims to providers (complex and uncertain in outcome), and support for organisations doing collective advocacy such as Artists Alliance or the Human Artistry Campaign.

Platform-specific policies

The landscape is shifting quickly, but as of today:

  • Spotify / Apple Music: require declaration of AI-generated content in the distribution process. AI music without disclosure can be taken down and the account suspended.
  • Instagram / Meta: prohibits deepfakes and misleading content. Realistic AI-generated images depicting real people without their consent carry risk of removal and account suspension.
  • Bandcamp: policy still evolving. As of 2025, it allows AI content if explicitly declared in the product description.
  • Galleries and art fairs: many have adopted policies excluding AI-generated work, or requiring mandatory disclosure. Read each open call’s terms carefully — most have added a specific clause in the past two years.

Practical recommendations

The legal framework is uncertain and evolving fast, but some principles hold regardless of how the rules shift:

  • Always disclose AI use when it is significant. Transparency is safer than concealment, and it builds trust with your audience, galleries and collaborators.
  • Read the terms of service of the tools you use, particularly the section on ownership of generated content. Some services retain broad licences over what you produce.
  • Document your process for work you intend to protect or submit to open calls.
  • For music, check your distributor’s policy (DistroKid, TuneCore, CD Baby) before publishing. Each has its own criteria, and conditions changed in both 2024 and 2025.
  • Follow the legal evolution: the EU AI Act will roll out in phases through 2027, and each phase may introduce new obligations or rights.

What is clear is that this issue is not going away. Artists who understand the legal landscape — even an imperfect and unsettled one — will be better positioned to make informed decisions about whether and how they incorporate AI into their practice.

If you have questions about how AI use affects your professional identity as an artist, write to us — we’re not lawyers, but we can help you understand the landscape and make decisions that are coherent with your trajectory.

Related articles: