On October 30 2024, the District Court of Amsterdam decided on a very interesting case about copyright infringement of RSS feeds, and I believe that some of the legal issues involved could prove to be relevant to open legal issues surrounding AI and copyright, particularly about the application of the text and data mining (TDM) exception in the EU. The case is DPG v HowardsHome (full text of the English translation here), and it involves European media conglomerate DPG and its Dutch subsidiary Mediahaus, which operates some regional and national publications; the defendant is HowardsHome, a commercial news aggregation service that provides news alerts of Dutch news.
HowardsHome had been providing its news alerts for years, processing up to 30,000 links per day on its website from RSS feeds. The service allowed users to customise feeds with various settings. HowardsHome primarily operated under an agreement that only applied if the full content of the linked articles was reproduced. In 2019, this agreement was terminated, and DPG announced that all future interactions with its content would be managed through Lexis Nexis. Subsequently, HowardsHome entered into an agreement with Lexis, both as a provider and to publish links and snippets from news articles.
Meanwhile, the claimants launched a company named ArtikelPro, which sold licensing agreements to small enterprises, seemingly in direct competition with the defendants. In March 2022, Lexis unilaterally terminated its licensing agreement with the defendants, arguing that HowardsHome was not only providing links sourced under its licence agreement with Lexis but also offering links outside the agreement. HowardsHome attempted to negotiate a new licensing deal, but no agreement was reached. DPG then filed a lawsuit asserting three claims: infringement of the publisher’s right, copyright, and database right.
While I’m sure many readers of this blog will be familiar with the publishers’ right (sometimes referred to as ‘the link tax’), I’ll nonetheless provide a brief introduction. The European publishers’ right was established in 2019 under the Digital Single Market and grants publishers of press publications the exclusive right to authorise or prohibit the online use of their content by digital platforms. This includes the use of snippets, headlines, and thumbnails, ensuring publishers receive fair remuneration for the sharing of their works. However, the right exempts individual users, hyperlinks, and the use of “very short extracts” to maintain a balance between accessibility and innovation.
The first issue considered in the decision was whether there had been an infringement of the publishers’ right. The defendants argued that they fell under the exemption to the publishers’ right, as they provided hyperlinks and short snippets not exceeding 150 characters. The court agreed, referencing recital 58 of the DSM for interpretation. This recital states that “it is possible that the use of individual words or very short fragments of press publications by information society service providers does not undermine the investment by publishers in the production of content.” Consequently, the court found that HowardsHome had not infringed the publishers’ right.
But it is the next stage of the decision that could prove to be interesting for copyright and AI. Besides the publishers right infringement, the claimants also alleged copyright infringement, they argued that in order to generate alerts from RSS feeds, the defendants had reproduced their copyright works. The defendants answered this point using several defences, I won’t go into all of them, but just highlight the relevant ones for AI. Firstly the defendants argued that not all articles published by DPG’s publications had copyright, the court had a short discussion citing Infopaq , but finally decided that while a determination of which articles had protection or not had to be done individually, there was enough evidence that at least a large number of publications had copyright.
Then the defendants argued that any reproduction that had taken place was temporary in nature, specifically trying to use the transient copy exception under Art 5(1) of the InfoSoc Directive. The court cited Infopaq II in this regard, where the CJEU set out 5 requirements for the transient copy exception to be used. The court argued that HowardsHome failed with regards to one requirement, which is that the copies cannot have independent economic value. The court states that “HowardsHome’s reproductions are central to its service: the reproductions are offered to its clients as alerts. Thus, these reproductions hold independent economic value and must be considered reproductions of the Publishers’ RSS feeds.” This could prove to be important as the transient copy exception may be used in future AI copyright cases, but now it has been denied both in this case, as well as in the recent LAION case in Germany.
But perhaps the most eye-catching element of this decision is that the defendants argued that any reproduction they may have conducted from an RSS feed was covered by the text and data mining exception that was introduced by Arts 3 and 4 of the DSM. Specifically, Art 4 covers the making of a reproduction of a work accessed legitimately for the purpose of extracting “patterns, trends, and correlations”. In an interesting turn of events, the court found that the defendants actions did indeed comply with this exception, and in doing so produced several interesting points that serve to elucidate the application of this exception.
Firstly, the court found that making a reproduction of content that is available on the public web through RSS (and not behind a paywall) can be considered to be a legitimate use in accordance to the exception, and HowardsHome can make copies of text to analyse it.
Secondly, the court specified the nature of the reservation of rights that is required (the opt-out in Art 4 DSM), and it specified that it has to be done in machine-readable format. This is important because the court in the LAION case had considered in an obiter dictum that the reservation of rights could be made in plain text. Moreover, the court made an interesting finding regarding the nature of the opt-outs made by the claimants, they had indeed included robots.txt file in their websites, but these only contained opt-outs against AI agents. I found this part of the decision incredibly ironic.
Finally, the court analysed whether the TDM exception complied with the three-step test. This was something else that was done in the LAION case, and here the court also ruled that the exception complied with this requirement under Berne. The decision states:
“HowardsHome argues that it meets the three-step test requirements. As explained extensively, HowardsHome’s service generates traffic to the Publishers’ websites, as clients are directed to the Publishers’ sites if they wish to read the full article. Moreover, HowardsHome asserts that a license for its alert service is not available, as the Publishers impose unrealistic conditions and excessive prices. The legitimate interests of the Publishers are not harmed by HowardsHome. For the third step, not all of the rights holder’s interests are relevant; only legitimate interests are considered. In this case, a balance must be struck between the copyright in the press publications and the freedom of expression in directing users to these press publications through HowardsHome’s alert service. The Publishers’ legitimate interest does not extend so far as to monopolize this flow of information.”
There was also a discussion regarding the database rights, but I will not be covering that in detail. Suffice it to say, the court also dismissed the argument that the defendants were infringing those rights. Given all of the above, the court found in favour of the defendants, and the claimants have to pay the costs of the proceedings.
Conclusion
While this is not an AI case, it is interesting how TDM cases are starting to make their way through the European courts. It is very interesting that at least at this early stage, the TDM exception has been successfully deployed in two different jurisdictions. There has been a lot of talk in some circles that has attempted to dismiss this exception, or attempted to reduce its applicability. At least for now, it seems like the TDM exception is alive and well, particularly after the AI Act reinforced its application by repeatedly referring to it. We are surely going to continue seeing cases across the EU where it will be applied by various courts. Next stop the CJEU?
The one thing that warms my heart is that RSS is still alive in the year 2024. As a long-time provider of an RSS feed, I’m happy that this technology is still relevant.
RSS, an elegant weapon for a more civilized age.
5 Comments
Uma · December 2, 2024 at 8:42 am
Really interesting case. Can I ask whether this is official translation or Google translation of case?
Andres Guadamuz · December 2, 2024 at 9:52 am
It’s not an official translation.
Paul Keller · December 2, 2024 at 9:00 am
Andres, I think you got the role of RSS in the case wrong. In this case it was HH that read RSS feeds made available by the Publishers. See section 2.3 of the ruling. So the RSS feeds are the source data that is being mined by HH and not the product (which is a subscription service). Does not change much about your analysis, but makes the Publishers position look even more unreasonable (why publish structured data if you do not want that data to be used by others?)
Andres Guadamuz · December 2, 2024 at 9:39 am
Thanks Paul! I understood that the feeds were copied by HH, but I also assumed the news alerts were in RSS format, I’ve amended, surprisingly little needed to be changed.
Anonymous · December 3, 2024 at 6:29 pm
What is the other case where the TDM exception has been successfully deployed?