The Patentability of Artificial Intelligence Inventions: the turning point of the UK Supreme Court and its possible implications in the European context

The Patentability of Artificial Intelligence Inventions: the turning point of the UK Supreme Court and its possible implications in the European context
On 11 February 2026, the Supreme Court of the United Kingdom delivered a judgment destined to leave a profound mark on international patent law. In Emotional Perception AI Ltd v Comptroller General of Patents, Designs and Trademarks, the Court held that an invention based on artificial neural networks (ANNs)[1], that is an artificial intelligence system, may, in principle, be patentable. The Court found that the exclusion from patentability of computer programs “as such” does not apply where the invention employs real physical hardware as an essential element of the claim; accordingly, it is not correct to regard an ANN as merely abstract software ipso facto excluded from patentability.

The appeal was brought by Emotional Perception AI Ltd, which, after successfully challenging before the High Court the rejection of its patent application by the Hearing Officer of the UK Intellectual Property Office, contested the subsequent judgment of the Court of Appeal. The latter, on appeal by the Comptroller General of Patents, had reinstated the original decision refusing the application. All decisions were made using the interpretative criteria from Aerotel Ltd v Telco Holdings Ltd (2006), a key English case on software patents.

The appeal concerned the possibility of obtaining, in the United Kingdom, the grant of a patent for a system based on an ANN. In particular, the appellant had filed a patent application for a system or method aimed at using an ANN to generate recommendations of media files, enabling a user interested in a given content - such as a musical track, video, or text - to receive suggestions of files capable of eliciting an emotional response analogous to that evoked by the original content. The appellant argued that the proposed system could provide this service more efficiently and accurately than solutions available on the market.

The dispute centred on the interpretation of Article 52(2)(c) and (3) of the European Patent Convention (EPC), as well as the corresponding provision in section 1(2)(c) of the Patents Act 1977, which excludes “programs for computers as such” from patentability. The appeal before the Supreme Court raised three fundamental issues: whether the criteria developed in Aerotel should be departed from; whether an artificial neural network could qualify as a “computer program” or contain one; and, finally, whether the overall subject matter of the claims should be regarded as wholly excluded from patentability.

In summary, on these principal issues the Court held, first, that UK courts should no longer apply the four-step test established in Aerotel Ltd v Telco Holdings Ltd & Ors to determine whether an invention, particularly software and business methods, is excluded from patentability, but should instead adopt the “any hardware” approach endorsed by the Enlarged Board of Appeal of the European Patent Office (decision G 1/19).

Secondly, the Court held that a neural network “constitutes a set of instructions for manipulating data in a specific manner in order to achieve a desired result” and must therefore be regarded as a “computer program” within the meaning of Article 52(2)(c) EPC.

As to the third and final issue, however, the Court did not consider it appropriate to decide on the substantive patentability of the claimed invention, since the case had been argued exclusively on the basis of the Aerotel approach, and the Court had not received sufficiently detailed submissions to apply the approach developed in the aforementioned case law of the Enlarged Board of Appeal of the European Patent Office. The Supreme Court therefore remitted the case to the Hearing Officer of the UK Intellectual Property Office to assess the remaining requirements for patentability of the claimed invention.

The decision marks a clear break with established English practice, which had been based on the Aerotel test developed in Aerotel Ltd v Telco Holdings Ltd (2006), and instead aligns UK law with the approach followed by the European Patent Office (EPO), as set out in the more recent decision G 1/19 of the Enlarged Board of Appeal. According to that approach, the presence of any hardware element confers a technical character on the claim capable of overcoming the exclusion and satisfying the initial admissibility threshold.

Although formally limited to the jurisdiction of the United Kingdom, the Supreme Court’s decision is of international significance, as it harmonises national practice with established interpretative principles at the European level. The relevant legal framework is provided by Article 52(2)(c) and (3) EPC, which, while excluding “programs for computers,” specifies that where the claimed subject matter possesses technical character, it is not excluded from patentability as such. The entire interpretative framework developed by the EPO is based on these provisions and aims to distinguish abstract software (which is not patentable) from software that produces a “further technical effect.” According to the consistent case law of the EPO Enlarged Board of Appeal, the exclusion from patentability is overcome where the execution of software on a computer produces effects that go beyond the normal physical interactions between software and hardware. For patentability purposes, according to established EPO case law, software must also provide a technical contribution over the prior art, must not consist merely of a method capable of being performed as a mental act, and as noted, must produce a further technical effect.

The EPO approach, often referred to as the “any hardware approach” or the “two-hurdle approach,” therefore entails a dual assessment. The first stage examines whether the claimed subject matter possesses technical character. The presence of any hardware element or interaction with an external physical process is generally sufficient to overcome this initial hurdle, thereby conferring on the claim the status of an invention within the meaning of Article 52 EPC. The second stage involves examining the invention considering the traditional patentability requirements of novelty, inventive step, and industrial applicability.

The UK judgment thus adopts, in a decisive departure from domestic case law applied over the past twenty years, a principle already firmly established within the European patent system, according to which the focus of the analysis lies not on the nature of the medium (software), but on the technical result achieved.

It should be emphasised that the Supreme Court’s decision in Emotional Perception AI Ltd v Comptroller General of Patents does not open the door to indiscriminate patentability of any AI technology. While the judgment has been welcomed by UK practitioners as encouraging patent protection for AI innovations, it merely clarifies that such inventions are not excluded a priori, and they must still satisfy the rigorous traditional requirements of novelty, inventive step, and industrial applicability. Furthermore, the issue of software patentability remains distinct from the separate and equally debated question of whether artificial intelligence systems may be designated as inventors. On that point, the UK Supreme Court itself, in Thaler v Comptroller-General, held that an artificial intelligence system cannot be named as an inventor since the law requires the inventor to be a natural person.

In conclusion, the recent UK judgment represents an important step toward international convergence in the protection of innovations in the fields of software and artificial intelligence, as well as toward the harmonisation of the relevant legal framework at the international level.

Lawyer Arianna Serafini



[1] An Artificial Neural Network (ANN) is a model inspired by the functioning of the human brain, capable of learning through machine learning techniques. It can be trained to perform a specific task and to progressively improve its performance through an iterative process involving comparison between the generated outputs and predefined target outputs.

Newsletter

Iscriviti per ricevere i nostri aggiornamenti

* campi obbligatori