CJEU receives first referral on chatbots and copyright

Over the weekend, reports circulated that the Court of Justice of the European Union (CJEU) has received its very first referral on Artificial Intelligence (AI) and copyright.

Like Company v Google, C-250/25 comes from Hungary. It concerns chatbots vis-á-vis the press publishers’ right and the text and data mining (TDM) provision under, respectively, Articles 15 and 4 of the DSM Directive.

The referral, in essence, is asking the CJEU to clarify the applicability of the rights of reproduction and making available to the public in relation to the reproduction and display, in the responses of Google’s chatbot, of content that is partially identical to the content found on a newspaper’s website.

Legal context

As EU-based readers will know, Article 15 of the DSM Directive grants press publishers a related right over their press publications, which consists of controlling their reproduction and making available to the public by ISPs. The scope of protection under that right is equivalent to the rights found under Article 2 and 3(2) of the InfoSoc Directive.

In turn, Article 4 of the DSM Directive mandates upon EU Member States to introduce an exception or limitation to the rights of extraction and reproduction for the purpose of TDM. Crucially, the provision does not mention other exclusive rights, including the right of communication/making available to the public. It is also subject to a reservation possibility, in accordance with paragraph 3.

Referred questions

The Budapest District Court (Budapest Környéki Törvényszék) is asking the CJEU to clarify the following issues:

1. Whether the display, in a chatbot, of content that is identical to the protected content found on a publisher’s website is an act of (a) reproduction and (b) making available to the public and, if so, whether it matters that a response provided by a chatbot is the result of a process in which the chatbot merely predicts the next word on the basis of observed patterns;
2. Whether AI training engages the right of reproduction; if so,
3. Whether the TDM exception in Article 4 of the DSM Directive applies.
   

Comment

The referred questions go to the heart of some of the most pressing issues currently facing copyright and AI.

While I have addressed them in greater detail in this recent study amongst others, I would offer a couple of remarks in this post.

1. Chatbots’ display of third-party content

Two things to note at the outset here:

• The first is that the Hungarian court has phrased the referral in such a way that it appears to have been already established that the content found on the press publisher’s website is protected. If that was not the case, it could be noted that protection – under both copyright and the press publishers’ related right – does not inter alia encompass “mere facts”, this being a potentially contentious issue when it comes to content of a journalistic nature.
• The second is that neither copyright nor the press publishers’ right protect against independent creation: derivation must be proved. Commentators have noted that this might be challenging in the context of LLMs, given that it may be unclear what content the training has been done on. In turn, that is the top reason why rightholders have been advocating – both in the context of the forthcoming Code of Practice under the AI Act and the ongoing UK (potential) reform debate – for robust transparency obligations on the side of AI developers.

Having pointed out the above, if the content in question is protected and derivation is demonstrated, then – under EU law – the answer appears to be that the operation of a chatbot does likely engage both the rights of reproduction and making available to the public.

By way of reminder, in its case law, the CJEU has consistently held that a high level of protection of intellectual property rights needs to be ensured. This is mandated by relevant directives – including the InfoSoc and DSM Directives – and also stems from Article 17(2) of the EU Charter. In turn, a broad construction of the scope of protection must be guaranteed.

As to the sub-question whether “prediction” makes a difference, the answer might be in the negative if the result of such a prediction is the reproduction of third-party protected content. Whether that is the case will obviously depend on factual circumstances: not every “match” will be actionable.

2. Whether AI training is relevant under the right of reproduction

Turning to the second issue, this appears to revisit the vexed question whether training is even relevant under copyright law. At least in Europe, the answer seems to have been given already: yes.

Not only has the EU adopted specific exceptions to copyright and other rights (sui generis database right and certain related rights) to allow, at certain conditions, TDM, but the CJEU has consistently construed the right of reproduction broadly, in line with the very wording of Article 2 of the InfoSoc Directive.

It is clear that if the right of reproduction was not engaged by TDM, which is part of but is not all that AI training consists of, then the very existence of the TDM exceptions would be unjustified if not altogether senseless.

3. Whether the TDM exceptions apply to AI training

The final issue is also central to the discussion of whether the EU legislature even anticipated the explosive rise of generative AI when it adopted its own TDM exceptions back in 2019.

While some commentators have also recently answered in the negative, it seems to me that this discussion is somewhat self-resolved.

Not only was AI in the picture at the time when the DSM Directive was drafted, but the adoption of the AI Act – including the link that it does establish between TDM and AI training – has now settled this matter for good.

That said, what is key to remember is – once more – that TDM is relevant to AI training but is not solely what AI training consists of: as of today, there is no exception under copyright and other rights that encompasses “AI training” as such. In the EU, there are exceptions to specified restricted acts for specified purposes, all subject to several different conditions including (i) lawful access to the protected content in question and (ii) the requirements of the three-step test.

Next steps

And, speaking of steps, what will happen now? The CJEU judgment in this referral will likely be issued sometime in 2027 (late 2026 at the very earliest). A bet? Plenty more will happen in the AI/copyright space before then. Stay tuned!

[Originally published on The IPKat on 26 May 2025]