Artificial intelligence is generating commercially valuable innovation across almost every sector of the economy. Whether it is a recommendation engine, a diagnostic system, an image classifier or a generative model, the technical capabilities of AI systems represent a genuine investment in research, engineering and data processing. The question of whether those systems, and the inventions embedded within them, can be protected by patents in the UK is therefore of significant commercial importance.

The answer, following the UK Supreme Court’s judgment in Emotional Perception AI Limited v Comptroller General of Patents [2026] UKSC 3, is: potentially and generally yes, but the outcome depends on the specific invention, how it is described and how it is claimed.

This guide explains what represents likely patentable innovation in, particularly, the field of AI innovation in the UK, what the current legal framework requires, and what the Emotional Perception ruling changed.

Can you patent an AI invention in the UK?

Yes, AI inventions can be patentable in the UK, but not simply because a system uses artificial intelligence; there needs to be something more. The relevant question is whether the claimed invention, viewed as a whole, has technical character – and really a technical effect – and includes features that contribute in a technical sense when assessed through the appropriate legal framework.

Before the Emotional Perception ruling of 11 February 2026, many AI and computer-implemented patent applications in the UK failed at the statutory exclusion stage. The long-standing approach – set out by the Court of Appeal in Aerotel – required applicants to demonstrate that the “contribution” of their invention – what it had actually added to human knowledge – was technical in nature and did not fall solely within the categories excluded by section 1(2) of the Patents Act 1977 (UKSC). That analysis consistently produced restrictive outcomes for AI, software and neural network claims, especially since the sense of the term “contribution” is backward looking.

The Supreme Court has now replaced that approach with one that aligns UK practice with the EPO. Under the new “any hardware” framework, the exclusion threshold is lower: a claim that involves any piece of hardware is not, on that basis alone, a claim to excluded subject matter “as such.” The terms “as such” are expressly used in the UKPA and, also, the corresponding provisions of Art.52(3) of the European Patent Convention (EPC). A more demanding and objective analysis now moves to the so-called “intermediate step” which is used to identify those features that contribute to the technical character of the invention when considered contextually “as a whole.” This requires that features that, at first sight, appear to be “non-technical” in nature must be considered in terms of their interactions with other technical features of the claim to deliver a technical effect.

For AI applicants, this change materially improves the prospects of obtaining UK patent protection, provided that the invention is properly framed and both the technical implementation and related technical effect of the system are clearly identified and claimed.

What types of AI invention can be patented in the UK?

In the UK, patentable AI inventions extend across a wide range of technologies. The common requirement is that the claimed subject matter must have technical character: it must involve technical means and deliver a effect. The label “AI” or “neural network” does not determine patentability either way.

Types of AI invention that may benefit from UK patent protection include:

Artificial neural networks with a defined technical application. An ANN trained to perform a specific function – processing sensor data, improving image resolution, analysing audio signals for technical anomalies – may be patentable if the training architecture, the feature interactions and the technical output are properly claimed. Although the UK Supreme Court confirmed that ANNs are considered to be “programs for computers” and therefore potentially within the computer program exclusion, the presence of hardware (which is always required to implement an ANN) takes the claim outside the “as such” exclusion and into an assessment where structure, training, delivered function and surrounding technical implementation detail must be considered.

Machine learning systems embedded in technical processes. A machine learning model that forms part of a larger technical system – for example, a trained classifier used in industrial quality control, an ML-based signal processing pipeline or a predictive model integrated into a control system – may be patentable where the technical interactions between the model and other system components are clearly identified and claimed. This may reduce even further to the point where the ANN’s learning objectives are initially defined.

Recommendation and search systems using semantic or vector-based methods. The Emotional Perception invention itself concerned a recommendation system using ANN-based vector comparison. The Supreme Court’s ruling that this invention is not excluded from patentability creates a clear precedent for similar technologies operating in the space of embedding-based search and retrieval. Although an EPO Technical Board of Appeal case, reference is made to T1249/22, especially sections 10 and 11.

AI-enabled data processing and analysis systems. Systems that use trained AI models to process sensor data, analyse medical images, detect anomalies in industrial processes or classify information may be patentable where the AI component contributes to a technical solution. One should bear in mind that the term “technical effect” is forward thinking, whereas the term “technical contribution” appears to represent a backwards thinking analysis. The presence of an identifiable technical effect is therefore a critical success factor.

Hybrid products combining learned models with hardware. Where a trained model is embedded in a physical product – a medical device, an autonomous vehicle subsystem or an industrial controller – the combination of AI functionality and hardware implementation may support a strong patentability argument.

What is not patentable, even after Emotional Perception?

The UK Supreme Court’s ruling lowers the exclusion hurdle but does not remove it nor does it reduce the requirements for novelty and inventive step. Several categories of AI-related subject matter are likely to remain problematic.

An AI system whose sole claimed innovation is a business method, a mental act or a mathematical method – without any identifiable technical effect, interaction with the physical world or technical implementation that delivers a discernible technical advantage – is unlikely to satisfy the technical character requirement even under the “any hardware” approach. Including a computer in the claim does not automatically give the invention technical character; it is the function performed and the technical effect that matters.

AI systems that replicate known technical processes without a genuine inventive step face the same obstacles as any other claimed invention: the combination of features, including the AI components, must not be obvious to a person skilled in the art at the relevant date. Simple automation of an existing process is unlikely to yield patentable invention. However, the devil is in the detail, and that requires complete dissection and appreciation of the real-world impact.

Claims that are drafted too broadly or too abstractly, failing to identify the specific technical interactions that make the invention work, are likely to struggle at the intermediate step analysis, even if they pass the “any hardware” threshold, or later under inventive step considerations.

What changed for AI patents after the Emotional Perception ruling?

The most significant practical change is the abandonment of the Aerotel approach to statutory exclusion. Under Aerotel, patent examiners at the UKIPO were required to identify the “actual contribution” of the invention, what the inventor had really added to human knowledge, and to ask whether that contribution fell solely within excluded subject matter. This framework made it very difficult for AI and software applicants because examiners could characterise the contribution of any AI system as essentially a program, a mathematical method or a mental act, and refuse the application without engaging properly with the technical means through which that contribution was achieved. 

Note the earlier commentary on the meaning of: the term “contribute” as used Headnote 1 of T641/00-Comvik, i.e., an inclusive term; the term “contribution” as used in Aerotel and the backward facing nature of this term in the context of which it was used, and the term “technical effect” which is causal in nature and forward thinking enquiry.

The “any hardware” approach does not ask what the contribution is. It asks whether the claim involves hardware. If it does, the exclusion does not apply “as such.” The question of what the invention actually accomplishes technically – and whether it is novel and inventive – is assessed at the next stage under the proper and objective consideration of the intermediate step and, particularly, the novelty and inventive step frameworks. The point is that, as the WTO’s Art.27(1) TRIPS recites, patent protection is available in “all fields of technology.”

This change has several practical consequences. First, fewer applications should fail at the exclusion stage, and search results should therefore be supplied instead of a boilerplate rejection and refusal (as frequently occurred by applying Aerotel). Second, the intermediate step analysis, which asks which features contribute to the technical character of the invention when considered “as a whole,” becomes the critical analytical battleground between the exclusion threshold and the inventive step assessment. Third, the framing and quality of the patent specification matters more than ever because it is the foundation on which the intermediate step analysis and the inventive step assessment will be conducted.

How should AI patent claims be drafted after Emotional Perception?

Effective AI patent claim drafting after Emotional Perception requires attention to several key principles.

The Holistic Character Test: Identify what is being achieved in a technical sense. 

Claims should be grounded in the context of an articulated technical problem. This serves a purpose for other countries or regions, such as the EPO. Presenting a solution to the problem is also important, but one must understand that the UK does not adopt problem-solution as the process by which inventive step/obviousness is assessed, and problems can be framed as aims to be achieved in a non-technical field. The point is that the solution must inherently have a technical effect, and that will require some form a technical implementation that delivers an advantage.

Setting up the invention in the context of a specific limitation of the prior art is an effective way to start. This can take many forms, including the technical challenges faced by the system and how these are overcome. The claim needs to recite the features that together interact provide a technical effect in the context of the invention considered “as a whole.” Generic descriptions of AI functionality are less useful than claims that specify the architecture, training methodology and technical interactions that make the invention work.

Identify and claim feature interactions. The intermediate step requires identification of which features contribute to the technical character of the invention “as a whole.” Claims that fail to capture the interactions between the AI component and other system elements – hardware, data structures, processing pipelines and non-technical features – make it harder to argue that the full combination of features supports patentability. Specification drafting should explain not just what each feature does in isolation, but how the features interact to produce the technical effect.

Functional language. Claims that describe an AI system entirely in terms of what it does, rather than how it achieves that function, give examiners less to work with when assessing the technical character of the invention. This is attempting to claim an invention as a result to be achieved, which is permissible, but more difficult. The more clearly the technical mechanism is identified and claimed, the stronger the basis for arguing that the invention satisfies the inventive step and that the feature interactions – under the intermediate step – require proper consideration. There are instances where a product-by-process represents a viable approach, with this requiring that the product is new, its process is new and the function it delivers is new.

Consider claim hierarchy carefully. Independent claims should capture the core technical contribution at an appropriate level of generality. Dependent claims should develop the architecture, training approach and system integration at increasing levels of specificity, preserving fallback positions for prosecution.

How does the UK now compare with the EPO for AI patents?

The Emotional Perception ruling significantly closes the gap between UK and EPO practice for AI and computer-implemented inventions. The UK now adopts the same “any hardware” interpretation of Art.52(2) and (3) EPC, and the intermediate step analysis follows the same framework used by the EPO under the T641/00-Comvik and T154/04-Duns Licensing principles, as expressed in G1/19.

One area where UK and EPO practice continues to diverge is the inventive step methodology. The EPO still prefers the problem-and-solution approach as its standard method for assessing inventive step, but it is also not the only approach used by the EPO. UK courts apply the Pozzoli test, which remains valid under the Emotional Perception ruling albeit modified to includes steps (A) to (F) of Duns Licensing and rejects problem-solution. The Pozzoli questions:

(1)(a) Identify the notional “person skilled in the art”.

(1)(b) Identify the relevant common general knowledge of that person.

(2) Identify the inventive concept of the claim in question or if that cannot readily be done, construe it.

(3) Identify what, if any, differences exist between the matter cited as forming part of the “state of the art” and the inventive concept of the claim or the claim as construed.

(4) Viewed without any knowledge of the alleged invention as claimed, do those differences constitute steps which would have been obvious to the person skilled in the art or do they require any degree of invention?

The convergence in exclusion analysis means that the strategy developed for the EPO, particularly in relation to technical character and feature interaction, is now more directly transferable to the UK position.

Hepworth Browne’s AI patentability and computer-implemented invention services support clients with both UK and EPO strategies. If you have a pending UK application or are considering a new AI filing, you can also explore the firm’s guidance on AI and computer program patents.

Frequently asked questions: AI patents in the UK

Does using AI in a product mean the product can be patented? Not automatically. The product or process must have a technical character, be novel and involve an inventive step. The presence of AI is a factor in the analysis but not a substitute for it. The situation post-Emotional Perception is much improved, if not more favourable in the UK than anywhere internationally.

Is there a specific legal definition of “AI” in UK patent law? No. Patent law does not define AI. For patentability, the analysis focuses on the technical character of the claimed invention, its features, their interactions and the technical problem they address.

Can a trained AI model itself be patented? A trained model is an ANN with frozen weights and biases. The Supreme Court held that such a model is a “program for a computer” and therefore falls within the computer program exclusion “as such,” but if it is implemented by or with any technical means then this exclusion does not take effect. The finding, however, remains technically questionable. ANNs, realised as an emulation, are certainly computer implemented and make use of code and mathematics during training, but there is a deliverable end product. Expressing this slightly differently, whether the system or method implementing the model can be patented depends on whether the wider claim, including hardware and system components, involves more than the excluded matter “as such.”

How long does it take to get an AI patent in the UK? The UKIPO statutory compliance period is approximately four and a half years under standard procedures. However, these time lines can be shrunk considerably to see rights potentially allowed/secured within about 12 months from filing. This is significantly quicker than obtaining a UK patent through the EPO route, and there is also no centralised opposition attack against a UK patent. Following the Emotional Perception ruling, there may be strategic reasons to prioritise UK direct filing and prosecution alongside EPO prosecution.

What should I do if my AI patent application was refused in the UK? You may have grounds to revisit the application in light of the change in legal framework. However, this depends on precise timing.  Read our detailed guidance on reviving refused AI patent applications or contact Hepworth Browne for a review of the application.