[Guest post] The patentability of prohibited practices in the field of AI

Katfriends Andreas Engel (Heidelberg University) and David Faber (Gramm, Lins & Partner) introduce to the readers of the IPKat the key arguments of their article recently published in ZGE/IPJ (open access, in German, with further references) on the patentability of AI practices prohibited under Article 5 EU AI Act. They write:

"The EU AI Act in its Art. 5 prohibits certain AI practices. The pertinent recital explains that these practices contradict essential Union values. This sounds reminiscent of the rationale of exclusions to patentability under European and German law, Art. 53(a) EPC and Section 2 German Patent Act (Patentgesetz – PatG), which refer to ‘ordre public’ and morality. Still, we argue that there is no blanket exclusion of the patentability of inventions relating to AI practices prohibited by the AI Act.

Patentability of AI technologies within the scope of the AI Act in general

The AI Act uses a broad definition of AI systems; its focus is on machine-based systems that use subsymbolic knowledge processing techniques through artificial neural networks (ANNs). In its details, the examination and granting practices of European and German patent authorities is heterogeneous (cf, e.g., here). In the present context, suffice it to note that inventions with regard to ANNs can be eligible for patent protection in accordance with Art. 52, 54 seqq EPC / Sec. 1, 3 seqq PatG.

Exceptions to patentability under European and German patent law in general

Under European and German patent law, patents are not granted for inventions “the commercial exploitation of which would be contrary to ‘ordre public’ or morality”, see Art. 53(a) EPC and Sec. 2(1) PatG. This exclusion – which has been met with criticism by scholars – can be explained with the consideration that patent authorities cannot by official act grant protection if the invention is contrary to fundamental legal and ethical principles.

At the same time, this exclusion is to be narrowly construed (see, e.g., here at 6.2.2). Even according to the wording of Art. 53(a) EPC and Sec. 2(1) PatG, the commercial exploitation of an invention “shall not be deemed to be so contrary merely because it is prohibited by law or regulation”. Hence, not every prohibition excludes pertinent inventions from patentability. In particular, a prohibition that does not cover every conceivable exploitation of an invention does not imply a general verdict that the exploitation of the invention would be illegal or immoral by the standards of Art. 53(a) EPC and Sec. 2(1) PatG. As the EPO held, risks must be carefully weighed against an invention’s usefulness (balancing approach).

Kat-centred 

AI Act prohibitions for certain AI practices and their link to provisions on patentability

Since February 2025, the EU AI Act has prohibited certain AI practices in its Art. 5. These prohibitions can, in principle, be relevant for patent exclusions, as the AI Act covers many use cases that would be considered a commercial exploitation within the meaning of patent law.

Art. 53(a) EPC and Sec. 2(1) PatG come into play, as the prohibitions in the AI Act are imbued with the same essence that characterises ordre public and morality: In explaining the foundations of the prohibitions, recital 28 AI Act refers to “respect for human dignity, freedom, equality, democracy and the rule of law, as well as [the] fundamental rights enshrined in the Charter”.

Breaking the link, part 1: Exceptions to the specific prohibitions of the AI Act

Still, the vast majority of prohibitions enumerated in Art. 5 AI Act encompass specific exceptions within the operative part of the respective provision – Art. 5(1)(d), (f), (g) and (h) AI Act – or in the respective recitals – Art. 5(1)(a), (b) and (c) AI Act and recitals 29 and 31 AI Act. Even though the extent of some of these exceptions remain uncertain, it is safe to say that these prohibitions are not comprehensive, but leave open the possibility of exploiting pertinent inventions in conformity with the law – and, for the purposes of the balancing approach, show that pertinent inventions can be useful. Hence, the aforementioned prohibitions do not allow the conclusion that the exploitation of corresponding inventions would always be contrary to ordre public or morality.

By contrast, there is no comparable specific restriction on the prohibition of AI systems “that create or expand databases for facial recognition by indiscriminately reading facial images from the internet or surveillance recordings” (Art. 5(1)(e) AI Act).

Breaking the link, part 2: General exceptions for (scientific) research and development

For these practices in particular, the AI Act’s general exceptions for (scientific) research and development (Art. 2(6) and (8) of the AI Act) have relevance. These exceptions limit the scope of the AI Act, even for practices that constitute commercial exploitations for the purposes of patent law. Thus, these exceptions, too, weigh against the assumption that the AI Act would automatically lead to exclusions from patentability for prohibited practices. These general exceptions for (scientific) research and development serve to promote innovation, which also would have to be considered under the EPO’s balancing approach.

Case-by-case examination still necessary

However, a case-by-case examination of patent filings regarding prohibited AI practices remains necessary. Such an examination has to take into account the basis of the intended use of the invention, as objectively apparent from the disclosure in the patent application. This is likely to be particularly relevant for inventions for which the patent application suggests an embodiment that is subject to a prohibition. At the same time, it is to be expected that patent applications will use disclaimers.

Conclusio

These findings conform not only with the general reservations of patent offices in applying exceptions to patentability, but also with the thrust of AI regulation in the EU: The AI Act employs a differentiated regulatory technique that is commensurate with its complex regulatory objective of promoting human-centred and trustworthy AI, and aims at not unnecessarily restricting the miscellaneous benefits of AI technologies. In this light, the restriction of certain practices of exploitation in the AI Act seems sufficient to enforce the judgements expressed in the prohibitions, without the need for an additional blanket exclusion from patentability."