Short answer
Yes. The sui generis database right is a distinct layer of protection that operates independently of copyright.
"The right provided for in paragraph 1 shall apply irrespective of the eligibility of that database for protection by copyright or by other rights. Moreover, it shall apply irrespective of eligibility of the contents of that database for protection by copyright or by other rights." —
That sentence is the answer. Facts are not copyrightable, but a collection of facts in which someone made substantial investment can be protected against extraction and re-utilisation.
Three layers can apply to the same dataset, and each must be analysed separately:
- Copyright in individual contents — the works inside the database.
- Copyright in the database structure — where selection or arrangement is the author's own intellectual creation.
- The sui generis right — protecting the maker's substantial investment in obtaining, verifying or presenting the contents.
A developer scraping a price list, a sports fixture list, a company register or a job board may face no copyright issue at all in the individual entries, and still infringe the third layer.
What the right protects
"Member States shall provide for a right for the maker of a database which shows that there has been qualitatively and/or quantitatively a substantial investment in either the obtaining, verification or presentation of the contents to prevent extraction and/or re-utilization of the whole or of a substantial part, evaluated qualitatively and/or quantitatively, of the contents of that database." —
Article 7(2) defines the restricted acts broadly:
- Extraction — "the permanent or temporary transfer of all or a substantial part of the contents of a database to another medium by any means or in any form".
- Re-utilisation — "any form of making available to the public all or a substantial part of the contents of a database by the distribution of copies, by renting, by on-line or other forms of transmission".
Scraping a database into a training corpus is a transfer to another medium. It is extraction on the face of the definition, whether temporary or permanent.
The investment must be in obtaining, not creating
The single most important limit comes from the 2004 quartet of judgments, of which The British Horseracing Board (C-203/02) is the best known, decided alongside Fixtures Marketing (C-338/02, C-46/02 and C-444/02).
The Court distinguished investment in obtaining existing data from investment in creating the data in the first place. Resources used to generate the data that makes up the contents do not count towards the substantial investment that triggers the right.
That distinction bites hard in practice:
- A sports body generating its own fixture list invests in creating data, not obtaining it.
- A price comparison service collecting prices published by others invests in obtaining.
- A registry receiving mandatory filings may be doing less obtaining than it appears.
- Investment in verification and presentation counts, so curation, cleaning and quality control can establish the right even where the data was self-generated.
For an AI developer, this means the right is not automatic. It has to be established by the claimant, and it is frequently contestable — but where curation investment is real, it is frequently made out.
"Substantial part" and the repeated-extraction trap
Article 7(1) covers extraction of the whole or a substantial part, evaluated qualitatively or quantitatively.
Article 7(5) closes the obvious workaround:
"The repeated and systematic extraction and/or re-utilization of insubstantial parts of the contents of the database implying acts which conflict with a normal exploitation of that database or which unreasonably prejudice the legitimate interests of the maker of the database shall not be permitted." —
Systematic crawling is precisely the behaviour Article 7(5) addresses. A crawler taking small amounts repeatedly, over time, across a whole site, cannot rely on each individual request being insubstantial.
In Apis-Hristovich (C-545/07) the Court examined the concepts of permanent and temporary transfer and how to assess transfers between databases. In Innoweb (C-202/12) it addressed a dedicated meta search engine that enabled users to search the whole content of a third-party database in real time, treating that as re-utilisation.
CV-Online Latvia (C-762/19) supplied a balancing element: the Court addressed a search engine indexing and reproducing job advertisements, and framed the question by reference to whether the act adversely affects the substantial investment in obtaining, verification or presentation — introducing a competition-sensitive assessment rather than a purely mechanical one.
The TDM exception applies here too
The DSM text-and-data-mining exceptions were drafted with database rights expressly in mind. Article 4(1) provides an exception to, among others, "Article 5(a) and Article 7(1) of Directive 96/9/EC", and Article 3(1) does the same for the research exception.
So the same framework governs:
- Article 3 — research organisations and cultural heritage institutions, lawful access, no opt-out, contractual override unenforceable under Article 7(1) DSM.
- Article 4 — everyone else, lawful access, subject to an express reservation under Article 4(3), retention only for as long as necessary for the mining.
Two practical consequences. A machine-readable reservation on a database removes the Article 4 exception for the sui generis right just as it does for copyright. And access terms matter for the lawful-access condition — terms of use prohibiting automated collection go directly to whether the material was lawfully accessible.
The Database Directive also contains its own user provisions. Article 8(1) permits a lawful user of a publicly available database to extract or re-utilise insubstantial parts for any purpose, while Article 8(2) prohibits acts conflicting with normal exploitation or unreasonably prejudicing the maker's legitimate interests. Article 9 permits Member States to allow extraction of a substantial part in limited cases, including for illustration for teaching or scientific research, non-commercial and with the source indicated.
Who holds the right, and where
The sui generis right belongs to the maker — the person who took the initiative and the risk of investment. Article 7(3) makes it transferable, assignable or licensable.
Importantly, the right is an EEA right with a qualification requirement tied to the maker's nationality or establishment. A database made by a US company may not attract the sui generis right at all, while an equivalent European database does. This produces asymmetries worth checking before assuming a dataset is protected.
The protection term is fifteen years from completion, with substantial new investment capable of starting a fresh term — which for continuously updated databases can be effectively perpetual in practice.
Practical guidance for AI developers
- Analyse three layers separately: content copyright, database copyright, sui generis right.
- Ask whether the claimed investment is in obtaining, verifying or presenting — not in creating the data.
- Do not rely on facts being unprotectable. Article 7(4) makes that irrelevant to the sui generis right.
- Treat systematic crawling as within Article 7(5), not as a series of insubstantial takings.
- Check access terms, which bear on lawful access under DSM Article 4(1).
- Detect and honour reservations on databases as well as on works.
- Check the maker's establishment, since non-EEA makers may not hold the right.
- Consider licensing for high-value curated datasets; the sui generis right is precisely what makes them licensable.
- Record your assessment — the AI Act Article 53(1)(c) copyright policy covers related rights, not only copyright.
Conclusion
"The data is just facts" is not a defence in the EU. The sui generis right was created to protect investment in collections, and Article 7(4) states in terms that it applies irrespective of whether the database or its contents attract copyright.
The right is not unlimited: investment in creating data does not count, the maker must qualify, and the TDM exceptions apply to the sui generis right as they do to copyright. But for a well-curated, verified, continuously maintained database held by an EEA maker, it is a real and separately enforceable obstacle — and systematic scraping is the conduct Article 7(5) was written for.
Sources
This article provides general legal information, not advice for a specific matter. Laws, facts and procedural deadlines may change the result. Follow the linked primary sources and obtain local advice before acting.