The Supreme Court's Landmark Decision in Emotional Perception AI v Comptroller General
On 11 February 2026, the UK Supreme Court handed down its judgment in Emotional Perception AI Ltd v Comptroller General of Patents, Designs and Trade Marks [2026] UKSC 3. It is the most significant UK patent decision in years, and for businesses developing AI and software technology, the message is clear: the UK just became a substantially more welcoming jurisdiction for patent protection.
For nearly two decades, applicants seeking patents for computer-implemented inventions in the UK have faced a framework that was, by common acknowledgment, more restrictive than its European counterpart. The Supreme Court has now dismantled that framework and replaced it with the approach used by the European Patent Office. The practical effect is that AI and software inventions will no longer be refused at the threshold simply for being "software." Instead, they will be assessed on their merits: their novelty and inventive step. That is where strong innovations can compete.
The Old Problem: Being Turned Away at the Door
Since 2006, UK patent examiners have applied a test known as the Aerotel approach (after the Court of Appeal decision in Aerotel Ltd v Telco Holdings Ltd [2006] EWCA Civ 1371). Under this framework, the examiner's task was to identify the "actual contribution" made by the invention and then ask whether that contribution fell within excluded subject matter. Categories like computer programs, mathematical methods, and business methods cannot be patented "as such" under the Patents Act.
In practice, this meant that if the core of what you had invented was a new algorithm, a new way of training a neural network, or a clever software architecture, the examiner could conclude that your "contribution" was essentially just software and refuse the application before ever considering whether your invention was genuinely novel or inventive. You were being turned away at the door.
This created a well-known divergence with the EPO. Applicants would frequently obtain a European patent for an AI or software invention, only to have the same invention refused by the UK Intellectual Property Office. For businesses planning their IP strategy, the UK became the awkward jurisdiction: the one that didn't quite work for software.
What the Supreme Court Has Changed
The Supreme Court has now overruled the Aerotel approach entirely and aligned UK law with the EPO's interpretation of the European Patent Convention. The shift has three key dimensions.
First, the threshold question is now much simpler. Instead of dissecting the "contribution" of the invention and asking whether it is essentially excluded matter, the first question is simply whether the invention involves any technical means, in practice any hardware. Does the claim refer to a computer, a processor, a storage device, a network? If yes, the invention clears the first hurdle. It qualifies as an "invention" within the meaning of the law, and the exclusions for computer programs, mathematical methods, and the like do not apply to block it at the threshold.
"A claim directed to a computer-implemented invention avoids exclusion under Art 52 EPC merely by referring to the use of a computer, a computer-readable storage medium or other technical means." (EPO Enlarged Board of Appeal, G1/19, endorsed by the UK Supreme Court)
Virtually every real-world AI or software system involves hardware of some kind. This means the vast majority of computer-implemented inventions will now pass the threshold test.
Second, your invention is assessed on its merits. Once the threshold is cleared, the examination moves to the substantive questions: is the invention novel, and does it involve an inventive step? This is a fundamentally fairer assessment for AI and software innovations, because it focuses on what the invention actually achieves rather than categorising it as "just software." A genuinely novel and non-obvious AI system, whether that is a new training method, a new architecture, or a new application of machine learning, now gets the chance to be evaluated on those terms.
Third, the UK and EPO now speak the same language. The Supreme Court was explicit that UK courts and the UK IPO should follow the EPO Enlarged Board's interpretation of the European Patent Convention unless convinced it is wrong. For businesses filing in both jurisdictions, this means less divergence, fewer surprises, and a more coherent IP strategy. The days of tailoring different arguments for the UK and EPO on the same invention should, in large part, be over.
AI and Neural Networks: Firmly in Scope
The case itself concerned an artificial neural network, the foundational technology behind most modern AI systems. Emotional Perception AI had developed an ANN-based recommendation system that could suggest media files (music, video, text) likely to elicit a similar emotional response to a file a user already liked, purely by analysing the physical properties of the files rather than relying on semantic descriptions.
The Supreme Court held that an ANN is, in law, a "program for a computer." This might sound like bad news, but it is not. The Court defined an ANN as an abstract model, essentially a set of instructions to manipulate data in a particular way, which can be implemented on any hardware from a conventional computer to dedicated neuromorphic chips. The entire ANN, including its topology, activation functions, weights, and biases, constitutes the program. This is true whether the ANN is implemented in software or embedded in hardware, and whether or not its parameters were set by a human programmer or arrived at through machine learning.
Importantly, being a "program for a computer" does not end the analysis. Under the new framework, an ANN implemented on hardware is not a computer program "as such". It has technical character, and so it qualifies as an invention. The exclusion is not triggered. The Court allowed Emotional Perception AI's appeal and sent the case back to be assessed on its substantive merits.
For AI businesses, this is the key takeaway: the technology at the heart of modern AI (neural networks, deep learning, machine learning systems) is not excluded from patent protection in the UK. The question is whether your specific implementation is novel and inventive, not whether it is "really just software."
The New Battleground: Technical Contribution
The Supreme Court was candid that it could not resolve everything. Having overruled Aerotel and lowered the threshold for qualifying as an invention, the Court identified what it called an "intermediate step," drawn from the EPO's case law, that will become the focal point of future examination.
The intermediate step sits between the threshold question ("is this an invention?") and the substantive assessment of novelty and inventive step. Its purpose is to filter out features of the invention that do not contribute to its technical character, so that only technically contributing features are considered when assessing whether the invention is genuinely novel and non-obvious.
This is a feature-by-feature analysis. A claim may contain a mix of technical features (hardware, data processing architecture) and non-technical features (a business rule, an aesthetic choice, an abstract mathematical concept). Non-technical features are not automatically excluded. They remain in the analysis if they interact with the technical features to produce a technical effect. It is only non-technical features that make no such contribution, non-technical features "as such," that are filtered out.
The Supreme Court deliberately left the detailed application of this intermediate step to be developed by the UK IPO and the lower courts, recognising that it had not received full argument on the point. This means there will be a period of uncertainty as the new framework is applied in practice. But the direction of travel is clear: the question is no longer "is your invention excluded?" but "which features of your invention contribute to its technical character, and are those features novel and inventive?"
What This Means in Practice
For AI and software businesses considering patent protection in the UK, several practical points emerge from this decision.
The barrier to entry is lower. Under the old framework, many computer-implemented inventions were refused before their novelty or inventive step was ever considered. That front-door rejection is now far less likely. If your invention involves any hardware (and it almost certainly does) it will be assessed on its substance.
Filing strategy should align with the EPO. The convergence between the UK and EPO means that a single, well-drafted patent specification can now be prosecuted more consistently across both jurisdictions. Applicants who previously avoided UK national filings for software or AI inventions because of the Aerotel hurdle should reconsider.
Claim drafting still matters. While the threshold is lower, the intermediate step means that features which do not contribute to the technical character of the invention will be disregarded when assessing inventive step. Claims should be drafted to clearly articulate the technical features of the invention and how they interact. A claim that is mostly non-technical with only a token reference to hardware will clear the threshold but may struggle at the inventive step stage if the genuinely inventive elements are filtered out as non-technical.
Machine learning innovations are on stronger ground. The Court's detailed analysis of ANNs, and its rejection of the argument that machine-learned parameters are somehow different from human-programmed instructions, provides a clear foundation for patenting innovations across the AI stack: from novel training methods to new network architectures to specific applications of machine learning.
What Hasn't Changed
It is important to be clear-eyed about what this decision does and does not do. The Supreme Court has not opened the floodgates to patenting any software. Novelty and inventive step remain rigorous, independent requirements. A patent application for an AI system that is technically competent but not genuinely novel, or that represents an obvious application of known techniques, will still be refused, and rightly so.
Purely abstract innovations, such as a new mathematical formula with no technical implementation or a business method with no technical effect, remain excluded. The "as such" qualifier means that the exclusions in the law still have teeth; they simply operate differently under the new framework.
There is also genuine uncertainty ahead. The intermediate step has not yet been applied in the UK, and it will take time and further cases for its contours to become clear. The Supreme Court acknowledged this openly, noting that it would be "unwise" to create binding precedent on the intermediate step without the benefit of in-depth argument and the views of the specialist courts.
Looking Ahead
Emotional Perception AI marks a turning point. For nearly twenty years, the UK's approach to patenting computer-implemented inventions has been out of step with the EPO, to the detriment of innovators seeking protection in this jurisdiction. That misalignment has now been corrected at the highest level.
The timing is significant. AI is advancing rapidly, and the UK's ability to offer robust, predictable patent protection for genuine innovations in this field is increasingly important for domestic businesses and international applicants alike. The Supreme Court's decision sends a clear signal: the UK is open for business when it comes to AI patents.
For businesses developing AI and software technology, the practical message is straightforward. If you have a genuine technical innovation, whether that is a novel training method, a new neural network architecture, or a creative application of machine learning that solves a real problem, the UK patent system is now better equipped to recognise and protect it. The door that was too often closed is now open. What matters is whether your invention is worth walking through it.