EMOTIONAL Perception AI Ltd. v Comptroller-General of Patents, Designs and Trade Marks [2026] UKSC 3, [2026] Bus. L.R. 411 (“Emotional Perception”) marks the UK Supreme Court’s first decision directly on the legislation governing what is eligible for patent protection in the UK. It is also an example of the Supreme Court grappling with the connection between law and AI, in this case artificial neural networks.
The case centred on Patents Act 1977, section 1 (“PA 1977”), which implements European Patent Convention 1973, Article 52 (“EPC 1973”). These provisions allow patents conferring 20 years’ market exclusivity to be granted for “inventions” only. They also require the invention to be new, involve an inventive step and be capable of industrial application. Thus, the meaning of the word “invention” is a threshold of considerable economic and policy significance, shaping incentives for innovation.
The alleged invention was claimed as a “system” containing an artificial neural network (“ANN”) trained on paired physical features and human emotional descriptors capable of providing improved recommendations of “semantically similar” media – such as audio, video, images or text files. Semantic similarity refers to whether items evoke similar meanings or emotions. For example, the system could recommend songs that evoke similar emotions – such as joy or melancholy – by analysing measurable properties of music, including tempo, pitch and rhythm. As well as the ANN, the system also included a collection of stored files (e.g. music), a communications network and a networked device (e.g. a phone or computer) for passing on the recommendations. Media recommendation systems have commercial applications and have attracted significant investment from technology companies, although the case did not dwell on these specific issues.
Historically, when the PA 1977 repealed the Patents Act 1949, a positive definition of an invention was replaced with a list of categories that were not to be regarded as inventions. PA 1977, section 1(2), implementing EPC 1973, Article 52(2), excluded, “among other things”:
(a) a discovery […] or mathematical method;
(b) a literary […] work […];
(c) […] a program for a computer;
(d) the presentation of information.
The legislation states that these exclusions apply only where the patent relates to such subject matter “as such” and the introductory phrase “among other things” suggested that the list was not closed.
The unusual drafting of PA 1977, section 1 and EPC 1973, Article 52 soon gave rise to interpretative difficulties, resulting in divergence between the UK courts and the European Patent Office (“EPO”). In Aerotel Ltd. v Telco Holdings Ltd. [2006] EWCA Civ 1371, [2007] R.P.C. 7, the Court of Appeal had adopted a four-step test to apply Article 52: (1) construe the claim; (2) identify the actual contribution to prior knowledge; (3) determine whether that contribution falls solely within the excluded subject matter; and (4) check whether the contribution is technical in nature. In contrast, the EPO had emphatically criticised this approach in T154/04, Duns Licensing Associates [2007] E.P.O.R. 10 and followed a different line of reasoning which was later upheld by the EPC’s Enlarged Board of Appeal in G1/19, Pedestrian Simulation [2021] E.P.O.R. 30 (“G1/19”). The EPO applied a more permissive “any technical character” test under which a claim is treated as an invention if it includes any technical feature, with further filtering during the inventive step analysis. To prevent patents having an inventive step resting solely in excluded subject matter, the EPO added an “intermediate step”, under which only features contributing to the technical character of the invention are considered for inventive step (T 641/00, Comvik v GSM AB [2004] E.P.O.R. 10 (“Comvik”)). The so-called Comvik “intermediate step” has no explicit legislative basis and has drawn criticism from UK courts. Although subsequent UK decisions had softened the vehement disagreement, differences persisted, creating uncertainty whether EPO patents designated in the UK might be revoked.
Against this background, the Supreme Court in Emotional Perception was asked whether Aerotel should be overruled. The Court answered affirmatively, allowed the appeal and declared the ANN claims in the patent to be eligible inventions under Article 52. In doing so, it significantly reduced the long-standing disagreement between UK courts and the EPO about the eligibility of computer programs and aligned UK law with the Enlarged Board of Appeal’s interpretation of EPC, Article 52 in G1/19. The Court also held that the Comvik “intermediate step” applied in the UK, although it declined to replace the domestic framework for inventive step analysis (Pozzoli SpA v BDMO S.A. [2007] EWCA Civ 588, [2007] F.S.R. 37) with the EPO’s problem–solution methodology and remitted the case back to the IPO’s Hearing Office to apply the intermediate step and inventive step doctrines.
Legal practitioners have largely welcomed the decision on Article 52. Its outcome is relatively clear, consistent with EPO jurisprudence used on a daily basis and imposes few restrictions on patentable subject matter. The persuasiveness of the reasoning, however, is questionable. The foundational reason for the Court’s decision was that UK courts should follow the jurisprudence of the Enlarged Board of Appeal (and established precedent from the Technical Boards of Appeal) unless convinced that it is wrong or beyond the bounds of reasonable disagreement (at [44], [58]). This appears to mean that in the event of reasonable disagreement between UK courts and established EPO case law, UK case law should give way. Such a strong principle of deference, for the purposes of harmonisation, could have significant implications for the future of UK patent law in situations well beyond PA 1977, section 1 and patent eligibility.
The Court ought to have purposively interpreted the relevant patent legislation more carefully, in line with fundamental principles that statutes should be interpreted in accordance with Parliament’s purpose (R. (on the Application of Quintavalle) v Human Fertilisation and Embryology Authority [2005] UKHL 28, [2005] 2 A.C. 561) and treaties in light of their object and purpose (Vienna Convention on the Law of Treaties, Article 31). Harmonisation with the EPC 1973 was undoubtedly an important aim of the PA 1977 (see PA 1977, section 130(7)), but as Lord Sales observed extrajudicially in 2025, statutes frequently pursue multiple purposes and courts should avoid assuming a single mischief (Lord Sales, “Purpose in Law and in Interpretation” (2025) 74 International and Comparative Law Quarterly 753). In this case, as well as harmonisation, the PA 1977 strove to modernise the patent system and ensure that patent protection was confined to subject matter that promotes innovation without unduly restricting competition. The meaning of computer program was briefly discussed in this way during the statute’s Second Reading (HC Deb. vol. 379 cols. 270, 301–2 (24 January 1977)).
A more faithful application of statutory and treaty interpretation would have revealed substantial reasons to doubt the EPO’s “any technical character” approach, both as a reading of EPC 1973, Article 52 (per the Vienna Convention on the Law of Treaties) and as one consistent with Parliament’s intention in enacting the PA 1977. There is little in the text, structure or apparent object of EPC 1973, Article 52 to support the view that its exclusions are organised around the absence of “technical character”, although this has become a cornerstone of EPO jurisprudence. The term “technical” does not appear in EPC 1973, Article 52 or elsewhere in the EPC (except when referencing a scientifically qualified examiner), nor do the travaux préparatoires indicate that the drafters intended a single overarching principle (J. Pila, The Requirement for an Invention in Patent Law (Oxford 2010)). Rather, the structure of EPC, Article 52 suggests that the excluded categories reflect several distinct policy rationales. Several UK courts and commentators have long understood them in this way, identifying purposes such as preventing socially disruptive monopolies, avoiding overlap with copyright and excluding abstract or mere representation of information from proprietary protection (e.g. CFPH LLC’s Patent Application [2005] EWHC 1589, [2006] R.P.C. 5, at [21], [30]–[33]; Aerotel, at [9]).
The strongest argument for the EPO’s “any technical character” approach being a faithful purposive interpretation of Article 52 flows from the revisions to the EPC 1973, made in 2000 (European Patent Convention, 10th rev. ed., (January 2000)). The phrase “in all fields of technology” was added to EPC 1973, Article 52, following a Basic Proposal (part of the travaux préparatoires) which said that this was “with a view to enshrining ‘technology’ in the basic provision of substantive European patent law, clearly defining the scope of the EPC and making it plain that patent protection is available to technical inventions of all kinds” (“Basic proposal for the revision of the EPC”, MR/2/00, (2000), 43). This would appear to mean that inventions must be a form of “technology”, which could be subtly different from the EPO’s precedents on “technical character”. However, the Basic Proposal contained a further paragraph stating that the new wording reinforced the position that “patent protection is reserved for creations in the technical field” and that “[i]n order to be patentable, the subject-matter claimed must therefore have a ‘technical character’ or to be more precise – involve a ‘technical teaching’” (ibid., at 48). Based on this passage, it is arguable that the EPO’s technical character approach became the purposive intention of all EPC Member States after the 2000 revisions to the EPC took effect in 2007. However, the UK Government neither refuted nor endorsed the statements in the travaux préparatoires to EPC 2000, Article 52 nor amended PA 1977, section 1 in implementing legislation in 2004 – leaving some doubt whether these rationales in the Basic Proposal were fully accepted. Furthermore, if the EPC 2000 revisions were an important legal source for the interpretation of PA 1977, section 1 and EPC 1973, Article 52, the Supreme Court could have said so.
By failing to engage with these considerations at greater length, the Court gave priority to harmonisation rather than determining whether the EPO’s approach was in fact a defensible interpretation of both the Convention and the statute that gives it domestic effect. To be clear, if the Supreme Court recognised that the statutory exclusions pursue several purposes, it would not have been compelled to choose the Aerotel four-step test. It could have concluded that neither Aerotel nor the EPO’s “any technical character” approach offers a genuinely purposive interpretation of PA 1977, section 1 and EPC 1973, Article 52. The G1/19 approach reflects a particular understanding that is somewhat detached from the ordinary linguistic meaning of the provision, lacks support in the EPC 1973’s travaux préparatoires and express language, reflects German patent law traditions and relies on an intermediate step with no legislative foundation. It finds purposive support only in a minor paragraph of the travaux préparatoires for the EPC 2000 revision which has not been favourably referred to by either UK courts or Parliament. Meanwhile, the Aerotel test fails to recognise that the list of exclusions in PA 1977, section 1(2) and EPC 1973, Article 52(2) is not closed. Its second step focuses on the invention’s “contribution” without clearly defining the relevant prior art, its third step assesses whether the invention falls solely within an excluded category without reference to the underlying purposes of those categories and its fourth step introduces a “technical” requirement that finds no explicit basis in the statutory text. The adoption of a more rigorous purposive interpretation of the relevant patent laws could therefore have resulted in the rejection of both approaches and developed a more coherent, statute-centred framework. Such a course would have disrupted settled professional practice; yet, as the Court itself recognised, habitual adherence to a mistaken approach does not redeem legal error (at [50]).
Additional issues raised by the judgment cannot be explored fully here. However, some connected with the argument above warrant mention. First, it is questionable whether the Supreme Court fully appreciated the singular focus of the EPO’s technical character test: it devoted considerable attention to defining the terms “computer” and “computer program” and to determining whether an ANN falls within these concepts, despite the fact that the “any technical character” test of “invention” under G1/19 does not ultimately turn on such classifications. Second, and in contrast, the Court made relatively limited effort to clarify “technical” and “technical character”, referencing the “any hardware” factor rather than listing various forms of technical character recognised by the EPO, such as technical effect, technical application and technical means. Third, the Court chose not to take an important opportunity at the end of its judgment when it reiterated the EPO’s intermediate step, yet remitted it to the IP Office without a concluded view, despite the parties having already addressed many relevant points through the former Aerotel steps. This was a missed opportunity to engage with the most important substantive issue of the case and a point that could have influenced emerging case law at the Unified Patent Court.
Open access
