No Copyright in Laws or Decisions

First, it’s crucial to clarify that the content of Finlex – Finnish laws, statutes, and official decisions – is not protected by copyright. Section 9 of the Finnish Copyright Act explicitly excludes such official materials from copyright protection. In Finland, as in many jurisdictions, you cannot claim copyright over the text of an Act of Parliament, decisions of official bodies, or a court judgment. These texts are outside of copyright protection by law because the law is public information, and everyone must be free to read and use it.

So, if the dispute were simply about feeding the texts of laws from Finlex to an AI, copyright law wouldn’t pose a barrier.

However, what about the way Finlex organizes or presents that content? This is where database rights come in.

Sui Generis Database Rights

Finlex is essentially a database, a structured collection of legal documents. Even though the individual laws and decisions have no copyright, the database as a whole could be protected by a sui generis database right. Under Section 49 of the Finnish Copyright Act (implementing the EU Database Directive (96/9/EC)), a person or entity that has made a database with a substantial investment has the exclusive right to control the extraction or re-utilization of substantial parts of that database.

This right was originally introduced to protect investments in the compilation and structuring of vast amounts of data. But what is crucial is that the right is actually narrowly framed: the database maker can exercise control over the entirety of the database as a whole, or its “substantial parts”. However, the right does not give the maker any control over the pieces of data in the database.

But if the Finnish government were to use the contents of the entire Finlex to train AI, surely that would amount to “substantial part”?

Most likely. But there’s more to it in the case law.

The CV-Online Case: Introducing the “Adverse Effect” Test

This brings us to the CJEU’s CV-Online Latvia v. Melons case (Case C-762/19).

In this case, a company (CV-Online) had a public database of job listings, and another company (Melons) ran a specialized search engine that indexed those listings and provided links to them. CV-Online sued, claiming that Melons was extracting and reusing substantial parts of its database without permission. The dispute made it to the European Court of Justice, which had to interpret the Database Directive – essentially clarifying what counts as prohibited “extraction” or “re-utilization” of a database.

While the Court agreed that Melons did use substantial portions of CV-Online’s database, it ended up concluding that this alone was not enough to prohibit CV-Online from running its search engine.

In fact, an “adverse effect on investment” test was introduced for database right infringement. The Court (echoing the Advocate General’s opinion) held that a database maker can only prevent extraction or reuse under the sui generis right if the act in question undermines the investment that went into creating the database. In other words, not every technical reuse is an infringement – it has to be something that poses a real risk to the database maker’s ability to recoup their investment (for example, by threatening their revenue streams or business model).

This “adverse effect” criterion is not explicitly written in the statute; it’s a pragmatic gloss by the Court. The rationale is to strike a balance: database rights shouldn’t stifle useful services and further innovation on the back of data if they don’t actually harm the database owner’s economic interests.

In CV-Online, the service provided by Melons was a job aggregator that linked back to the original site. The Court noted that CV-Online’s site was still the one doing the “heavy lifting” – users ultimately landed there, meaning CV-Online wasn’t losing out in a way that jeopardized its investment. The database right exists to prevent free-riding that cripples the incentive to invest. Rather, the right is not meant to grant absolute control over how others may use the database.

This standard is critical when evaluating claims against AI developers or legal tech platforms that might train machine learning models on structured legal data.

So, Can Edita Prohibit AI Training on Finlex or not?

Applying the CV-Online ruling here, any argument that training an AI model on Finlex data constitutes unlawful re-use must meet the adverse effect standard. This seems like a high bar.

Finlex is not monetized. It’s publicly funded and free to access. Historically, Edita’s role, while important for the development of a functional service that is Finlex, was not based on a revenue model endangered by AI reuse. The original investment made in organizing Finlex was not made with an aim for Edita to invest in AI development. Therefore, it can be argued that using Finlex data to train AI cannot reasonably be said to “undermine” the original investment in the database.

Conclusion: Database Rights Cannot Justify Blocking AI Innovation

In early 2025, Finlex was updated, and the restrictions on the data use were removed. This change reflects both legal compliance and policy alignment: Finlex is now structurally and legally open by design. However, there is an argument to be made that the restriction should not have been there to begin with.

Database rights cannot be used to prohibit the training of AI, unless such reuse causes real economic harm to the rights holder.

The combined force of Finnish and EU law, along with judicial interpretation in CV-Online, points to a clear conclusion: database rights cannot be used to prohibit the training of AI, unless such reuse causes real economic harm to the rights holder.

By ruling on the CV-Online Latvia case, the European Court of Justice made a move towards favouring open data over proprietary in quite a fundamental manner, which has been largely ignored by the field.

It is up to the future to demonstrate whether this status quo of the law will become reality, or whether the case will be forgotten and stumped by companies and organizations not recognizing the potential for the use of available databases.