The rule in the AI Act
Providers of general-purpose AI models must have a policy to comply with EU copyright law, and in particular identify and respect rights reservations made under Article 4(3) of the EU copyright directive (2019/790), also with state-of-the-art technologies (Article 53(1)(c)). The recitals say this applies to anyone placing such a model on the EU market, regardless of where the training took place.
Text and data mining opt-outs
The copyright directive has exceptions for text and data mining, but rights holders can reserve their rights against it, except for scientific research. The Commission’s Code of Practice describes the reservation as made in any appropriate manner, such as machine-readable means for content made publicly available online. Where a valid reservation exists, you need permission to use the content.
What the Code of Practice suggests
The voluntary GPAI Code of Practice has a copyright chapter with these measures: a copyright policy, reproducing only lawfully accessible content, respecting rights reservations when crawling (following robots.txt as described in RFC 9309 and other appropriate machine-readable protocols), reducing the risk of infringing outputs, and a contact point for complaints.
Norway
An EEA Joint Committee decision to add the copyright directive was adopted in December 2023 but had not entered into force when we checked, and we found no text and data mining rule in the Norwegian Copyright Act. See the EU AI Act in Norway.
Keep a record of sources
For each data source, note where it came from, under what licence, whether crawling respected robots.txt and other opt-outs, and the collection period. The same record feeds the public training summary. This guide is general information, not legal advice.