Decisions · Policy · Germany

Copyright and the training of generative AI models (Urheberrecht und Training generativer KI-Modelle)

Tim W. Dornis (Leibniz University Hannover) and Sebastian Stober (University of Magdeburg) · 2024 · Proposal · non-binding

Nomos, Recht und Digitalisierung vol. 19, 2024 (open access)

AI training and copyright

Can AI be trained on protected works?

Only with permission or pay

Training generative AI is copyright infringement that the EU text and data mining exception doesn't cover

Why hereCoded at the pole: its core holding or recommendation answers the question outright.

The model as a copy

Is a trained model a copy of its training works?

Closer to a copy of its training works

Training works are fully or partly replicated inside the model

Why hereCloser to 'a copy of its training works': the study finds protected works are 'fully or partly replicated inside AI models' and treats this as a general feature of generative models, not limited to rare memorisation.

What it says

Combining legal and computer science analysis, the study finds that generative models memorise parts of their training data, that training is not text and data mining in the sense of the EU exception, and that providers outside the EU are bound by EU copyright when their systems are offered in the EU.

The leading interdisciplinary study for rightholders; commissioned by the Initiative Urheberrecht.

Sources

This record

Data as of 2026-10-21.

Drafted with AI assistance and checked against the sources linked above; not yet reviewed by a lawyer. Methodology

Changes

  • 2026-10-20Also coded on the new question on the model as a copy.