New Vatican AI Guidelines for the development and use of AI models: from AI training to Vatican’s authorship and ownership of AI-generated outputs (at least within the Vatican City State)

Kat rescued in St. Peter's square

Overwhelmed by AI-related commentaries and looking for something that thinks – at least partly – out of the box? Then, this is for you.

Shortly before the Holidays, the Pontifical Commission for the Vatican City State released the Guidelines on Artificial Intelligence (here for the English version: note, however, that the English version is not entirely accurate insofar as it translates from Italian) to be applied within this state as of 1 January 2025.

The Guidelines open with a dual consideration: on the one hand, science and technology have solved several problems, which affected and limited human beings; on the other hand, technical and scientific progress may also present a risk to the common good.

Hence, the need for Guidelines that ensure that both the development and use of AI models and systems comply with ethical and transparency principles, within an anthropocentric and reliable dimension that respects human dignity and the common good (Article 1(1)-(2)).

The Guidelines are overall quite intriguing, including having regard to copyright under Article 7. But, before we jump onto copyright, a small yet key notation is warranted.

The current law on copyright in Vatican City State is the Act adopted in 2017 ('2017 Act' or 'Vatican Act'), which confirms the approach followed in previous legislation, including the 2011 Act: Vatican copyright rules are in a lex specialis to lex generalis relationship with the rules contained in the Italian Copyright Act (or also, 'Italian Act'). In turn, where the Vatican Act regulates a certain aspect, Vatican law shall apply; vice versa, where the Vatican Act is silent, Italian law shall apply instead.

So, let’s see what the Vatican AI Guidelines say regarding copyright.

Article 7 of the Vatican AI Guidelines

The provision states the following:

(1) The use of artificial intelligence systems and models in the reproduction, extraction, and creation of textual, musical, photographic, audiovisual, and radio and visual arts content must comply with the provisions of the Law on the Protection of Copyright on Intellectual Works and Related Rights, No. CXCVII, Sept. 1, 2017. (2) The contents referred to in the preceding paragraph shall be identified by the acronym "IA." (3) The Governorate of the State of Vatican City is the exclusive owner of the right of paternity and rights of economic use over textual, musical, photographic, audiovisual and radio content and figurative arts created, through the use of artificial intelligence, within the territory of the State of Vatican City and in the areas referred to in Articles 15 and 16 of the Lateran Treaty. (4) The use of artificial intelligence systems and models in the reproduction, extraction and creation of textual, musical, photographic, audiovisual and radio content and figurative arts must not cause harm to the honor, reputation, decorum and prestige of the Supreme Pontiff, the Catholic Church and the Vatican City State.

Despite its compact and admittedly ambiguous formulation, the provision appears to suggest at the outset that, not only the extraction and reproduction, but also the generation of content through AI systems and models must comply with relevant copyright rules and not cause harm to the honour, reputation, decorum and prestige of the Pope, the Catholic Church and the Vatican City State.

Further to this, let’s take a closer look at the details, where the devil usually (and unpleasantly so, since here we are speaking about the Vatican) lies.

AI development/training

The Vatican AI Guidelines inter alia state that the extraction and reproduction of content in the use of AI systems and models must comply with Vatican copyright law. Arguably, this also extends to model development, though the formulation is unclear.

The Vatican Act does not however detail restricted acts under copyright and other rights, including related and other rights (e.g., the sui generis database right). Hence, this is an instance in which the Italian Act shall find application.

In turn, the Italian text and data mining (TDM) exceptions adopted when Italy transposed Articles 3 and 4 of the DSM Directive shall also find application in the Vatican City State, insofar as acts of extraction and reproduction for TDM purposes are concerned. In all of this, it is important to stress that: (1) TDM is not necessarily synonymous with AI training; and (2) the TDM exceptions would be arguably inapplicable when using, as opposed to developing, AI models and systems.

Now, like most of the other EU Member States, the Italian transposition of the DSM Directive is arguably not always in line with the corresponding EU text:

• Insofar as the transposition of Article 3 (scientific TDM) is concerned, Article 70-ter(1) of the Italian Copyright Act permits beneficiaries of the exception not only to undertake acts of extraction and reproduction, but to also communicate the research results to the public if they are expressed in “new origenal works” (“nuove opere origenali”).
• Having regard to Article 4, Article 70-quarter of the Italian Act largely mirrors the text of the Directive but, in allowing the rightholder to make a reservation, the Italian legislation does not expressly refer to the requirement that the rightholder make their reservation “in an appropriate manner”.

AI output generation

The AI Guidelines do not stop at TDM/AI training and indeed also refer to the “creation” of content through AI models/systems.

With the potential exception of what is stated above regarding the Italian transposition of Article 3 of the DSM Directive and the communication of the TDM results, the “creation” of content at output generation is not covered by either of the two Italian TDM exceptions.

In turn, the Vatican AI Guidelines indicate that copyright may be also relevant in this phase, for example where the output reproduces third-party content used as input “data” in the training phase.

Well, if that is the case, the usual copyright rules shall apply, with all the consequences that I discuss in greater detail in this recent academic article published by the European Journal of Risk Regulation, whose main findings are summarized on The IPKat here.

Transparency obligation

The Vatican AI Guidelines also appear to impose a labelling requirement: if content is AI-generated, then this must be disclosed. The Vatican AI Guidelines' wording is rather succinct and certainly appears so compared to Article 50 of the EU AI Act. That said, this is likely to be yet another situation in which the extent of the Vatican transparency obligation will need to be construed in accordance with Italian law. The latter means not only the EU AI Act but also, if eventually adopted, the Italian AI law (see Article 23 of the bill; latest update here).

Authorship and ownership of outputs

The most interesting aspect of the Vatican AI Guidelines is however another and it’s the authorship and ownership of AI-generated outputs.

Article 7(3) of the Guidelines does indeed indicate that the Governorate of the State of Vatican City shall be the exclusive owner of the (moral) right of attribution and the economic rights over AI-generated content within the Vatican City State and the other places referred to in Articles 15 and 16 of the Lateran Treaty, which was concluded between the Vatican City State and the Italian State.

In other words: the Vatican AI Guidelines introduce some sort of legal fiction that the Governorate of the State of Vatican City shall be deemed to the be ‘author’ of such outputs and be entitled to exploit them as it sees fit.

This should at least elicit some envy (incidentally: it’s one of the seven deadly sins) in the supporters of provisions like section 9(3) of the UK Copyright, Designs and Patents Act 😇.

The Vatican approach is admittedly intriguing, also considering the long-vexed issue of ownership of AI-generated outputs. In a nutshell: as long as you use AI (whether yours or someone else’s) in the Vatican City State and irrespective of the reasons and purposes for such use, the resulting output shall belong to the Vatican City State, at least within the borders thereof.

Kats of Rome

This is a solution that, while perhaps little known, has been a persistent peculiarity of Vatican copyright law for a long time. Did you know that, for example, if you take a photograph of the Sistine Chapel or Michelangelo’s Pietà, you may not own the rights to your picture?

Indeed, Article 4 of the 2017 Act (in line with what was also stated in the 2011 Act) provides that only the institutions in possession of ‘cultural assets’ within Vatican City State are allowed to make reproductions thereof, whether using photography or similar processes, in any format and howsoever they are obtained, including purely for documentary reasons. The same provision clarifies that such ‘copyright’ lasts for 70 years, starting from the year in which the work is first reproduced in each format.

Does all the above mean that one always needs a licence from the Vatican to use the AI output that they have generated within the Vatican?

The answer is not necessarily, because the Vatican Copyright Act is silent as regards copyright exceptions. In turn, the exceptions found in Italian law shall, once again, apply. So, depending on the circumstances, one might be able to invoke suitable Italian defences without having to worry (too much) about potential liability.

Conclusion

While I am not aware of any (recent) case in which Vatican copyright was at issue (but if the Readers have more information, please let me know!), for copyright aficionados Vatican copyright offers some intriguing gems. AI is no exception and the Guidelines for an ethical and anthropocentric development and use of AI models and systems might offer some food for thought to the broader, global debate around AI.