UK Supreme Court broadens the circumstances in which computer programs and AI-related inventions can be patented in a ground-breaking ruling

The UK Supreme Court has brought the UK’s approach to the patentability of computer-implemented inventions into closer alignment with the European Patent Office, displacing the 2006 Aerotel framework and easing the threshold exclusion analysis for AI and software-related inventions. 

Yesterday’s judgment in Emotional Perception AI Limited v Comptroller General [2026] UKSC 3 affirmed the Court of Appeal’s conclusion that an artificial neural network (‘ANN’) constitutes a computer program for the purposes of section 1(2)(c) of the Patents Act 1977 (implementing Article 52(2) EPC). Claims directed to an ANN therefore engage the statutory exclusion for computer programs ‘as such’. However, the Supreme Court held that the UK’s Aerotel framework should no longer govern that assessment. Instead, the UKIPO and courts should apply an approach aligned with established EPO jurisprudence, including the ‘any hardware’ principle and the COMVIK methodology reflected in G 1/19.

In this article we briefly discuss the background of this case, the details of the approach now endorsed by the Supreme Court, and what that potentially means for patentability of AI and other computer-implemented inventions in the future.

Background

Emotional Perception AI applied for a UK patent for a system and method using an ANN to recommend media files to users based on emotional similarity said to derive from measurable properties of the files (for further information on this, see our earlier article here).

This application was initially rejected by the UK IPO, on the grounds of being a “computer program […] as such”, excluded subject matter under s 1(2)(c) of the Patents Act 1977.

This rejection was appealed to the High Court, which allowed the appeal, finding that an ANN is not a program for a computer. Alternatively, the High Court held that even if an ANN were a computer program, the subject matter of the invention was not excluded from patentability, on its application of the Aerotel test.

The Court of Appeal allowed the Comptroller’s appeal. It held that an ANN is a program for a computer, characterising the weights and biases of the ANN as instructions which cause a machine to process information in a particular way. It further concluded that the claimed invention did not make a relevant technical contribution under Aerotel.

The matter was then appealed to the Supreme Court.

Issue 1: What is a computer program?

To decide whether an ANN amounted to a computer program, the Supreme Court first addressed what constitutes a “computer”. It rejected the Court of Appeal’s broad definition of a computer as “a machine which processes information” as overly wide, encompassing devices such as ovens or analogue radios. Conversely, the applicant’s contention that “computer” should be limited to devices with CPUs was considered too narrow, as it would fail to accommodate technological developments such as quantum computing.

The Court concluded that a computer is properly characterised as a machine for performing computations in accordance with a set of instructions (a program) operating on data.

Applying this definition, the Court held that an ANN of the type claimed amounted to a computer program for the purposes of section 1(2)(c). An ANN is an abstract computational model—a network of artificial neurons applying mathematical operations to data—rather than a physical machine. Whether implemented on specialised hardware or on a conventional computer, it constitutes a set of computational rules or instructions causing the machine to process data in a particular way. The Court rejected the applicant’s attempt to distinguish between “hardware ANNs” and “software ANNs”, holding that the choice of implementation is a practical matter and not conceptually determinative for the purposes of the exclusion.

Issue 2: Was the application for a Computer Program “as such”?

A central issue before the Supreme Court became whether the established Aerotel framework should continue to govern the exclusion analysis, or whether the UK should adopt an approach aligned with EPO jurisprudence, including the Enlarged Board’s decision G 1/19 (which we previously covered here).

The Court emphasised the importance of consistency with EPO case law, observing that decisions of the Enlarged Board should ordinarily be followed unless there is good reason to depart from them. It accepted that Aerotel risked conflating the question of whether subject matter is excluded with the separate requirements of novelty and inventive step.

As a result, the Court indicated that assessment of computer-implemented inventions should proceed in a manner consistent with the EPO’s “any hardware” principle and the COMVIK approach. In practical terms:

Step 1: Does the claim involve any hardware? If the claim involves technical means (such as a computer or other hardware), it is not excluded merely because it relates to a computer program. 

Step 2 (intermediate step): Which features contribute to the invention's technical character? A key part of this analysis is to identify, as part of the inventive step assessment, which features of the claim contribute to the invention’s technical character. Features that do not contribute, such as those describing a business scheme or a mathematical method in the abstract, cannot support inventive step. However, a feature that is non-technical when viewed in isolation may still be taken into account if, in the context of the invention as a whole, it interacts with the technical subject matter to help produce a technical effect. In this way, the analysis focuses on the technical contribution made by the combination of features, ensuring that only those elements capable of supporting patentability are relied upon when assessing inventive step.

Step 3: Proceed to review novelty and inventive step for the remaining features. 

In the present case, the Court considered that the claims involved technical means and therefore were not excluded at the threshold stage. However, it declined to carry out the inventive step analysis under this revised approach as these matters had not been argued by the parties and the analysis would benefit from the specialist expertise of the UKIPO. The Court remitted the matter to the UKIPO for determination.

Comment

This decision strikes a careful balance between the broader patentability of AI proposed by the High Court and the narrower approach taken by the Court of Appeal. By adopting an approach aligned with the EPO’s “any hardware” methodology, the Supreme Court has eased the exclusion threshold for computer-implemented inventions. This should enable significant technical innovations in the rapidly developing field of AI to be protected, while the filtering of non-technical features ensures that purely abstract algorithms or business methods remain unpatentable.

Importantly, the Supreme Court confirmed that the UK’s established approach to assessing inventive step under Pozzoli remains valid and need not be replaced by the EPO’s problem-and-solution method. However, the requirement to identify which features contribute to technical character as part of the inventive step analysis will need to be integrated into existing UK practice. Some refinement may be required as the UKIPO and courts work through how this interacts with the identification of the inventive concept under Pozzoli.

The substantial body of EPO case law on computer-implemented inventions should provide helpful guidance for the UKIPO in implementing this revised approach.