A website being publicly visible is not itself a complete copyright clearance for AI training. European text-and-data-mining exceptions have different beneficiaries, conditions and rights-reservation mechanisms.
Research exception under Article 3
Directive (EU) 2019/790 Article 3 concerns reproductions and extractions by research organizations and cultural heritage institutions for scientific research, involving works to which they have lawful access. Copies can be retained with appropriate security for verification. A commercial lab should not simply assume the scientific-research exception applies because it studies models.
General mining exception under Article 4
Article 4 provides a broader exception or limitation for mining lawfully accessible material, but it is subject to rights being expressly reserved in an appropriate manner. The directive mentions machine-readable means for material publicly available online. Licensing, national implementing laws and the facts of access must still be examined.
Separate mining from publication of outputs
An exception relevant to reproductions during analysis does not automatically settle obligations involving output copying, trademarks, privacy or distribution of source datasets. Reproducing a long protected passage is a different scenario from learning statistical patterns from text. The applicable legal framework can change by member state and use.
Build a practical evidence file
Capture source URL, access route, time of collection, contractual access restrictions, technical rights reservations and any downstream rights claimed. Review the exact national implementation and relevant case law before production. Because robots rules and rights statements may change, periodically recheck permissions for new collection.
Example and limit
A European company wants to mine a licensed journal collection. It verifies lawful access, reads license terms, identifies any reserved mining uses and assesses whether its activity falls within Article 3 or 4. It does not claim that every online article is free for commercial training.
Sources & Further Reading
- Directive (EU) 2019/790 Articles 3 and 4 text and data mining — European Union EUR-Lex.
- Copyright and Generative AI Training: Creative Commons Issue Brief — Creative Commons.
- EDPB Opinion 28/2024: Data protection aspects of AI models — European Data Protection Board.
Evidence checked 25 October 2026. Applicability: European Union; national transposition matters. This material is general education, not legal, medical, investment or other professional advice. No vendor performance test is implied.