Artificial intelligence systems create intellectual property law problems at two distinct points in the AI lifecycle: at the input stage, when AI systems are trained on large datasets that may include third-party copyright-protected works; and at the output stage, when AI systems generate content, inventions, or designs whose ownership and protectability under existing IP law is unclear or contested.
Both issues are actively litigated and regulated, and both have practical consequences for organisations that develop or use AI systems commercially.
The EU AI Act addresses training data transparency specifically. General-purpose AI model providers must publish a summary of the copyrighted training data used to train their models and must comply with EU copyright law, including the text and data mining exception under the Copyright in the Digital Single Market Directive (CDSM Directive) and the rights-holder opt-out mechanism introduced by that Directive. This means that the IP dimension of AI is no longer a purely private law matter; it is now directly integrated into the AI Act's compliance framework.
Training a large AI model involves processing enormous quantities of text, images, code, audio, and other content. Much of this content is protected by copyright held by its authors, publishers, or other rights holders. The legal question is whether this use of copyright-protected content to train AI models requires a licence from the rights holders, or whether it falls within an exception to copyright infringement.
In the EU, the CDSM Directive provides a text and data mining (TDM) exception that permits the reproduction and extraction of lawfully accessed content for the purpose of TDM, including for AI training. However, the Directive also permits rights holders to opt out of this exception for commercial TDM purposes, by making the reservation of rights available in a machine-readable format. Rights holders who have exercised this opt-out (and many major content producers have done so) must be compensated or their content must not be used for commercial AI training without a licence. GPAI model providers subject to the AI Act must implement policies to comply with this opt-out mechanism.
Outside the EU, the position varies. In the US, ongoing litigation, including cases brought by newspapers, authors, and visual artists against major AI developers, is testing the boundaries of fair use as applied to AI training. The outcomes of these cases will shape the legal landscape for AI training data globally, but there is no settled answer as of 2025.
EU copyright law requires that a work be an original creation reflecting the author's own intellectual creation in order to attract copyright protection. This standard requires human authorship. Purely AI-generated content, without meaningful human creative contribution in the creation or selection process, does not meet this standard and is not copyright-protected. It is in the public domain and can be used by anyone.
The practical implication is significant for organisations that use AI to generate content commercially. A company that uses a generative AI system to produce marketing copy, product descriptions, software code, or design assets may not own copyright in those outputs, depending on the level of human creative input involved. The EUIPO and national courts across the EU have consistently applied the originality requirement to deny copyright protection to outputs that are primarily generated by AI without substantial human creative direction. Organisations relying on copyright to protect AI-generated content commercially should take legal advice on whether their specific use case involves sufficient human authorship to meet the originality threshold.
European patent law requires that an invention be made by a human inventor. The European Patent Convention's requirement that a patent application designate the inventor has been consistently interpreted to require a natural person. The European Patent Office has rejected applications that designate AI systems as inventors, and no EU member state currently grants patents for AI-invented subject matter without a human inventor on record. This means that technical advances generated primarily by AI (new molecules identified by generative chemistry models, software architectures produced by AI design tools, manufacturing process improvements discovered by AI optimisation systems) may not be patentable unless a human inventor can be identified as having contributed meaningfully to the inventive concept.
It depends on the level of your human creative contribution. Under EU copyright law, the originality standard requires that the work reflect your own intellectual creation: your personal creative choices in its expression. If your contribution consisted of writing a detailed prompt and then selecting, editing, and arranging the AI's output, there may be sufficient human authorship in the selection and arrangement. If you simply submitted a generic prompt and accepted the AI's output without modification, there is unlikely to be copyright protection. The question is highly fact-specific, and legal advice is recommended for any commercially significant AI-generated output.
For EU GPAI model providers subject to the AI Act, compliance with copyright law and opt-out mechanisms is a legal obligation, and they must provide a publicly available summary of training data. For organisations using GPAI models in their products or services, the vendor's training data compliance is a risk that should be addressed in the procurement contract. Indemnification against third-party IP claims arising from the vendor's training data is a standard provision to request. The AI Act's transparency requirements for GPAI providers make it easier to assess this risk than was previously possible.
Where a human inventor meaningfully contributed to the inventive concept (for example, by designing the AI's search parameters, interpreting its outputs, and recognising their inventive significance) that human can be named as inventor and a patent application can proceed in the normal way. The AI's role in generating the technical result does not disqualify the human inventor, provided that a genuine human creative or inventive contribution can be identified. What is not permissible is naming the AI as inventor, or filing a patent application that designates no human inventor because the invention was generated entirely by the AI without human inventive contribution.
%402x.svg){kind=icon}
%402x.svg){kind=small}
%402x.svg){kind=icon}
