Copyright in AI output: who owns AI-generated content?

Short answer: Under EU copyright law, protection arises only in an "author's own intellectual creation" by a natural person. Output produced purely by an AI system — without sufficient human creative input — therefore in principle enjoys no copyright protection. No one "owns" such content by virtue of copyright. Anyone who wants certainty over ownership and use must arrange it by contract.

The core: copyright requires a human

EU copyright protects works that are the expression of the author's own intellectual creation. The Court of Justice has repeatedly confirmed this human-creativity requirement. A machine is not an author, and the EU has no separate category for "computer-generated works". Without human creative choices there is therefore no work and no rights holder.

When does protection arise anyway?

The line lies in the degree of human input. A single, generic prompt ("make an image of a sunset") is generally insufficient: the creative choices then sit with the system, not the human. But as the human substantially steers, selects, edits and arranges the result, that human contribution can attract a protected work. What is protected is then the human creative contribution, not "the AI output" as such.

This is a gradual, fact-specific judgement. For businesses it means uncertainty: whether an asset is protected depends on the concrete creative process and is rarely guaranteed up front.

Settle ownership by contract

Because copyright in pure AI output is uncertain, practice shifts to the contract. The terms of use of the AI platform often determine who gets which rights and usage powers over the output. Watch for:

• What the provider terms say about ownership, exclusivity and commercial use of output;
• Whether the provider may reuse your input or output for training — see AI, trade secrets and confidentiality;
• Indemnity: does the provider cover third-party claims over infringing output?

The infringement risk from the other side

Beyond whether your output is protected lies the reverse: AI output can infringe someone else's rights. A model trained on protected material can produce results that resemble existing works too closely. That risk lies partly with the provider and partly with you as the user who publishes the output. This is why provider indemnity is not a detail but an essential part of the contract.

Transparency under the AI Act

Separate from copyright, the AI Act (Art. 50) requires that certain AI-generated or manipulated content be recognisable as such — including machine-readable marking of synthetic content and clear labelling of deepfakes. This concerns not the ownership but the recognisability of AI output.

What to do

• Do not assume automatic protection of pure AI output; treat it as potentially freely available.
• Document human creative input for assets in which you want to claim rights.
• Read the provider terms on ownership, reuse and indemnity before productive use.
• Capture it in policy: include rules on AI output in your AI use policy for employees.
• Mark synthetic content where the AI Act requires it.