Emotional Perception AI Ltd v Comptroller General of Patents [2026] UKSC 3: How the Supreme Court Rewrote the Rules on AI Patents

For twenty years, a UK patent applicant whose invention ran on a computer faced a particular kind of frustration. It was not enough to show that the invention was new, clever, or commercially valuable. The applicant first had to survive a threshold test that had nothing to do with those qualities. Under the Aerotel framework, the courts asked a preliminary question that tripped up a significant number of genuinely novel software and artificial intelligence inventions before they could even be assessed on their merits. In February 2026, the UK Supreme Court looked at that framework, found it logically flawed, and replaced it with a simpler, more rational approach. The result is one of the most practically significant intellectual property decisions of the decade.

This article explains what the old law required, what Emotional Perception AI actually built, how the case travelled through the courts, and what the Supreme Court’s new approach means for AI inventors and students of intellectual property law.

The statutory foundation: what does the Patents Act actually say?

A patent gives its owner the exclusive right to prevent others from using an invention for a limited period, typically twenty years. In return, the inventor discloses how the invention works, adding to the public pool of knowledge. The foundational conditions for patentability in the United Kingdom are set out in the Patents Act 1977. An invention must be new, involve an inventive step, be capable of industrial application, and must not fall within a list of excluded categories.¹

The exclusion relevant to software and artificial intelligence appears in section 1(2)(c) of the Act, which provides that a “scheme, rule or method for performing a mental act, playing a game or doing business, or a program for a computer” is not regarded as an invention.² The critical qualification is the phrase that follows immediately: the exclusion applies only to the extent that a patent or application “relates to that subject matter as such.”³

Those two words, “as such,” carry an enormous amount of weight. They signal that Parliament did not intend to exclude every invention that happens to involve a computer program. The question has always been: what distinguishes an excluded computer program “as such” from a patentable invention that is merely implemented by a computer program? The answer to that question has been contested in UK courts for decades, and it is precisely the question that Emotional Perception AI finally resolved.

The Aerotel problem

Before the Supreme Court’s 2026 ruling, the leading UK authority was Aerotel Ltd v Telkom Holdings Ltd; Macrossan’s Patent Application.⁴ The Court of Appeal in that case set out a four-step test. Courts were to construe the claim, identify the actual contribution the claimed invention made, ask whether that contribution fell solely within the excluded subject matter, and check whether it was actually technical in nature.⁵

On paper, this looks structured and fair. In practice, it created serious problems, especially for AI and software patents. The test required courts to identify the “contribution” of the invention at a preliminary stage, before properly examining whether the invention was novel or involved an inventive step. Critics argued, and the Supreme Court ultimately agreed, that this jumbled the logical order of analysis. The threshold question of whether something is even an “invention” was being answered by looking at its contribution, which is really a question about novelty and inventive step. The cart was being put before the horse.

The European Patent Office (EPO) took a different approach. Rather than asking courts to assess a contribution at the door, the EPO applied what practitioners came to call the “any hardware” test. If a claim involves any technical means, any physical component, it has technical character and clears the threshold. The harder questions about what is genuinely inventive are then dealt with properly at the inventive step stage.⁶ The UK and the EPO were applying fundamentally different frameworks to the same types of applications, and the divergence had grown increasingly untenable.

What Emotional Perception AI actually built

Understanding the case requires a brief, non-technical explanation of what an Artificial Neural Network is and what the company’s invention did.

A conventional computer program works by following explicit instructions written by a human programmer. If a programmer wants a system to recommend music, they write rules: if the user likes fast-tempo tracks, suggest other fast-tempo tracks. The logic is transparent and human-authored. An Artificial Neural Network works differently. Instead of following pre-written rules, it learns from data. It is fed a large quantity of examples and adjusts its internal numerical parameters, called weights, in response to patterns in that data. The resulting system produces outputs not because a human wrote instructions telling it to, but because it has learned statistical relationships from experience.

Emotional Perception AI developed a system using an ANN to recommend media files, such as music tracks, to users.⁷ The ANN did not rely on human-assigned categories or tags. Instead, it analysed the objective, measurable physical properties of the files themselves, including qualities like tempo, tone, and rhythm, and identified which files were closest to each other in a multi-dimensional mathematical space.⁸ When a user expressed a preference, the system would recommend files that occupied a proximate position in that space, without a human ever having decided what “similar” should mean.

The UK Intellectual Property Office (UKIPO) refused the patent application. Its position was that the ANN was a program for a computer and that the invention therefore fell within the section 1(2)(c) exclusion.

The journey through the courts

At first instance in the High Court, Sir Anthony Mann took a different view.⁹ His reasoning was that an Artificial Neural Network is not, in any meaningful sense, a computer program at all. A program consists of explicit instructions; an ANN operates through learned weights that no human ever wrote. On that analysis, the section 1(2)(c) exclusion simply did not apply, and the application should proceed to be assessed on its merits.

The Court of Appeal disagreed and reversed that decision.¹⁰ Applying the Aerotel test, the Court of Appeal reinstated the UKIPO’s refusal. In the Court of Appeal’s view, the ANN was properly characterised as a computer program and the application fell within the excluded category.

The case then reached the UK Supreme Court, which heard the appeal and handed down a unanimous judgment on 11 February 2026.¹¹ The five justices were Lord Briggs, Lord Hamblen, Lord Leggatt, Lord Stephens, and Lord Kitchin.

The Supreme Court’s reasoning

The Supreme Court’s starting point was to consider whether an ANN is a computer program. On this question, the Court departed from Sir Anthony Mann and agreed with the Court of Appeal: an ANN does constitute a computer program.¹² The learned weights that encode the network’s knowledge are the product of a computational process, and the system that runs those weights to produce outputs is, in the relevant sense, a program capable of being followed by a computer to produce data manipulations.¹³

However, the Court held that characterising the ANN as a computer program did not end the analysis, and this is where the judgment has its real bite. The question is not simply whether a computer program is involved. The question is whether the invention relates to a computer program “as such.” Under the “any hardware” test that the Supreme Court adopted from the EPO, an invention that involves any technical means possesses technical character and is not excluded.¹⁴

In the case before them, the invention intrinsically relied on hardware: a database that stored the media files and their learned relationships, a communications network over which recommendations were delivered, and a user device on which the application operated.¹⁵ Those hardware elements were not incidental to the invention. They were constitutive of it. The claim therefore possessed the necessary technical character, cleared the threshold exclusion, and was to be remitted to the UKIPO Hearing Officer for a substantive assessment of novelty and inventive step.

The Court also explicitly overruled the Aerotel test.¹⁶ Its criticism was pointed: Aerotel had reversed the proper logical sequence of analysis. By requiring courts to assess the “contribution” of an invention at the threshold stage, the test was effectively asking them to answer questions about novelty and inventive step before those questions were properly before them. The EPO’s approach, by contrast, keeps the threshold question simple and deals with the harder substantive questions where they belong.

The COMVIK follow-on: where the real battle now happens

Adopting the “any hardware” test does not mean that every computer-related application will now receive a patent. It means that far fewer applications will fail at the threshold stage. The substantive assessment of inventive step then becomes the primary battleground.

At that stage, UK patent practice now follows the COMVIK approach derived from EPO jurisprudence.¹⁷ Under this approach, when assessing whether an invention involves an inventive step, features of the claim that are not technical in nature are stripped out of the analysis and not counted toward inventiveness. A business method implemented on a computer, for example, would not gain an inventive step from the non-technical business logic. Only the genuinely technical aspects of the invention, those involving hardware, physical processes, or technical problem-solving, contribute to the inventive step assessment. This preserves the integrity of the patent system: inventions succeed or fail on their genuine technical merits rather than on how their claims are drafted.

What this means for AI inventors and students

The practical consequence of this ruling is significant. Before Emotional Perception AI, a substantial number of AI and software patent applications in the UK were rejected at the threshold stage, never reaching a proper assessment of novelty or inventive step. That threshold, governed by Aerotel, operated as a high and sometimes arbitrary filter. The Supreme Court’s adoption of the “any hardware” test dramatically lowers that initial barrier.

For an AI system to clear the threshold under the new approach, it needs to involve at least one physical component: a processor, a database, a network, a sensor, a device. In practice, this is not a difficult condition to satisfy. Almost every deployed AI system operates on hardware of some kind. The substantive question of whether the technical aspects of the system involve a genuine inventive step will remain difficult and will rightly filter out obvious or trivial claims. But applicants will at least reach that question.

The judgment also resolves a damaging divergence between UK and EPO practice. An applicant whose AI invention was granted a European Patent through the EPO faced the paradox of being refused a UK patent for the same invention under the Aerotel framework. That incoherence is now corrected. The alignment also has international significance: Singapore’s Intellectual Property Office launched a consultation in April 2026 to consider whether to adopt the UKSC’s reasoning, illustrating how influential the judgment has been beyond UK borders.¹⁸

The Nigerian perspective

Nigerian students reading this case should note that the Patents and Designs Act 1970, which remains the governing statute in Nigeria, broadly mirrors older British principles. An invention must be new, result from inventive activity, and be capable of industrial application.¹⁹ However, Nigerian patent law currently lacks any developed jurisprudence on the patentability of software or artificial intelligence systems, and no Nigerian court has yet grappled with how to handle applications involving ANNs or machine learning systems.

The African Union’s Continental AI Strategy, adopted in July 2024, urges member states to harmonise intellectual property frameworks to support AI innovation across the continent.²⁰ As Nigerian commercial courts and the UKIPO of Nigeria begin to encounter AI patent applications, they will face precisely the same interpretive question that divided UK courts for two decades: what separates a patentable AI invention from an excluded computer program? The Supreme Court’s reasoning in Emotional Perception AI offers a pragmatic, hardware-anchored answer that Nigerian courts and policymakers may find persuasive.

Key essay and exam arguments

Three analytical arguments are particularly strong for exam purposes.

The first concerns judicial activism and statutory interpretation. The Supreme Court’s adoption of the EPO’s “any hardware” test required it to depart substantially from two decades of domestic precedent. Students can critically evaluate whether this constituted legitimate purposive interpretation of the “as such” proviso, or whether the Court effectively rewrote the statutory framework to achieve policy alignment with the EPO. The stronger argument is that the Court’s approach is consistent with parliamentary intent: section 1(2)(c) was always qualified by “as such,” and the EPO test gives genuine content to that qualification in a way that Aerotel did not.

The second argument concerns the ontology of the ANN. The High Court and the Supreme Court reached opposite conclusions on whether an ANN is a computer program. Sir Anthony Mann’s reasoning, that a system operating through learned weights rather than explicit instructions is not a program in any meaningful sense, has doctrinal coherence even though the Supreme Court declined to follow it. Students should engage seriously with both positions and consider what it means for the law to treat a system that no human programmed as a “computer program.” The implications reach well beyond patent law into questions of authorship, liability, and agency.

The third argument concerns the shift in battleground. Adopting the “any hardware” test simply moves the primary question from subject-matter exclusion to inventive step. Students should assess whether this genuinely improves outcomes for meritorious AI inventors, or whether the COMVIK approach at the inventive step stage will prove equally restrictive by stripping out non-technical elements. The answer is likely that the change does improve outcomes, because the COMVIK framework at least allows the genuinely technical aspects of an AI system to be assessed properly, rather than being dismissed at a preliminary stage on a characterisation question.

Conclusion

Emotional Perception AI Ltd v Comptroller General of Patents is not simply a case about one company’s music recommendation system. It is the case that ended UK exceptionalism in software and AI patent law. By overruling Aerotel and aligning with the EPO’s “any hardware” test, the Supreme Court gave AI inventors a fairer path to protection while preserving the substantive rigour of the inventive step requirement. For students of intellectual property law, it is essential reading: a judgment that resolves a long-running doctrinal dispute, clarifies the relationship between UK and European patent practice, and asks enduring questions about what it means for the law to engage with technology it does not fully understand.

---

Footnotes

¹ Patents Act 1977, s 1(1).

² Patents Act 1977, s 1(2)(c).

³ ibid.

⁴ Aerotel Ltd v Telkom Holdings Ltd; Macrossan’s Patent Application [2006] EWCA Civ 1371, [2007] 1 All ER 225.

⁵ ibid [40] (Jacob LJ).

⁶ EPO Decision T 154/04 (Duns Licensing Associates LP) [2006]; EPO Enlarged Board Decision G 1/19 (Pedestrian Simulation) [2021].

⁷ Emotional Perception AI Ltd v Comptroller General of Patents, Designs and Trade Marks [2026] UKSC 3 [4]–[6].

⁸ ibid [5].

⁹ Emotional Perception AI Ltd v Comptroller General of Patents, Designs and Trade Marks [2023] EWHC 2948 (Ch) (Sir Anthony Mann).

¹⁰ Emotional Perception AI Ltd v Comptroller General of Patents, Designs and Trade Marks [2024] EWCA Civ 825.

¹¹ Emotional Perception AI Ltd v Comptroller General of Patents, Designs and Trade Marks [2026] UKSC 3.

¹² ibid [38].

¹³ ibid [36]–[38]; Bruce Dearling, ‘Emotional Perception AI Limited v Comptroller General: Analysis of the UK Supreme Court Judgment’ (Hepworth Browne, February 2026) https://hepworthbrowne.com/analysis-of-the-uk-supreme-court-judgement-in-emotional-perception-vs-comptroller-general-2026-uksc-3 accessed 9 June 2026.

¹⁴ [2026] UKSC 3 [42]–[44]; EPO Decision T 154/04 (n 6).

¹⁵ [2026] UKSC 3 [45].

¹⁶ ibid [46]; Aerotel (n 4) overruled.

¹⁷ EPO Decision T 641/00 (COMVIK) [2003]; [2026] UKSC 3 [48]–[50].

¹⁸ Baker McKenzie, ‘Singapore IPOS Consults on UKSC Emotional Perception AI Patent Decision’ (April 2026) https://www.bakermckenzie.com/en/insight/publications/2026/04/singapore-ipos-consults-on-uksc-emotional-perception-ai-patent-decision accessed 9 June 2026.

¹⁹ Patents and Designs Act 1970 (Nigeria), s 1(1).

²⁰ African Union, ‘Continental Artificial Intelligence Strategy’ (July 2024) https://au.int/sites/default/files/documents/44004-doc-EN-_Continental_AI_Strategy_July_2024.pdf accessed 9 June 2026.