UKIPO again refuses Emotional Perception AI patent after Supreme Court remittal

The UK Intellectual Property Office (UKIPO) has again refused Emotional Perception AI Limited’s patent application for an artificial neural network (ANN)-based media recommendation system, following the UK Supreme Court’s landmark ruling earlier this year that fundamentally changed the UK approach to patentability of computer-implemented inventions.

The decision, issued following the Supreme Court’s February 2026 judgment in Emotional Perception AI Ltd v. Comptroller General of Patents, Designs and Trade Marks [2026] UKSC 3, concludes that the application does not satisfy the requirements for patentability, including inventive step.

The Supreme Court had expressly rejected the UK’s long-standing Aerotel approach to the computer-program exclusion and directed the UKIPO to apply the framework established by the European Patent Office (EPO) Enlarged Board of Appeal in G1/19.

The latest UKIPO decision therefore provides the first detailed application of that new framework to the very invention that prompted the Supreme Court to change the law.

Supreme Court opened the door to patentability

Emotional Perception’s application, GB1904713.3, concerns a system using an ANN to generate media recommendations based on similarities between files.

The invention trains an ANN using two measures of similarity: a semantic distance derived from natural-language descriptions of files and a distance based on objectively measurable physical properties. Once trained, the ANN can use those physical properties to generate recommendations intended to produce a similar emotional response to an input file.

The application was originally refused by a UKIPO hearing officer in 2022 under the exclusion for “programs for computers … as such.” The High Court subsequently allowed an appeal, but the Court of Appeal reinstated the refusal in 2024.

The Supreme Court unanimously allowed Emotional Perception’s appeal in February 2026. It held that an ANN of the type at issue was a “program for a computer,” but concluded that the claims nevertheless had technical character because the ANN necessarily operated using hardware. The claims also referred to hardware elements including a database, communications network, and user device.

The Court therefore held that the claimed invention was not a computer program “as such.” It rejected the Aerotel methodology and adopted the approach in G1/19, while leaving the application of the so-called “intermediate step” and the question of inventive step to the UKIPO.

The Supreme Court emphasized that the intermediate step was intended to identify which features of a mixed technical and non-technical invention contribute to its technical character, before those features are assessed for novelty and inventive step. It specifically cautioned that the intermediate step should not simply reproduce the former Aerotel contribution analysis.

UKIPO reaches refusal on remittal

Following the Supreme Court judgment, the UKIPO reconsidered GB1904713.3 under the new framework. An examination report issued in March had already concluded that the claimed invention lacked novelty and/or inventive step. In particular, the examiner considered alternative ways of implementing the invention and concluded that the distinguishing features did not make a technical contribution capable of supporting an inventive step.

The examiner subsequently maintained the objections, leading to today’s hearing officer’s decision and renewed refusal of the application. The outcome means that, despite the Supreme Court’s finding that the claimed ANN system clears the “computer program as such” exclusion, the application has not cleared the separate patentability hurdle of inventive step.

The distinction is important. The Supreme Court did not find that the invention was itself patentable. Rather, it held that the UKIPO had applied the wrong legal test when it originally refused the application and remitted the case for consideration of the remaining requirements. The Court expressly left the application of the intermediate step and inventive step to the hearing officer.

Concerns over the “intermediate step”

Bruce Dearling of Hepworth Browne, who represented Emotional Perception AI in the litigation, has criticized the UKIPO’s approach.

“Any refusal is disappointing; this one doubly so,” Dearling said.

“There appears to have been no desire to actually look at the structure of the claimed invention and to apply the intermediate step. Rather, the UKIPO approach appears possibly to have been a deliberate shortcut to avoid having it actually make a decision on the approach dictated by G1/19 and the UKSC, maintaining the ‘unsound’ Aerotel contribution test in another guise.

“That is a worrying thought given that G1/19 is settled law that outlines what amounts to computer-implemented inventions, so is the UK now again in disharmony with the EPO and clearly settled jurisprudence that our judiciary has encouraged us to follow?

“Quoting Sir Robin Jacob in, ironically, Aerotel, ‘In saying our courts would and should normally follow the settled jurisprudence of the European Patent Office it should be understood, of course, that they are not bound to do so. In the unlikely event that we are convinced that the commodore is steering the convoy towards the rocks, we can steer our ship away.’

“Putting this in commercial context, Singapore is already looking to take on and run with broader AI and ANN protection whilst accepting most of the aspects of the UKSC decision. My immediate read of this decision is that the UKIPO has, in fact, steered us towards those very rocks that industry has sought to navigate around.”

Dearling’s comments are particularly significant because he was the attorney responsible for the case architecture before the UKIPO and the UK courts.

Why the decision matters for AI patents

The case has effectively produced a two-stage development in UK AI patent law.

First, the Supreme Court lowered the threshold for getting an AI or computer-implemented invention past the initial “invention“ hurdle. Under the approach it adopted from G1/19, the presence of hardware can be sufficient to establish technical character. The Court expressly said that the previous Aerotel contribution analysis was “unsound“ because it risked conflating the question of whether something was an invention with the separate questions of novelty and inventive step.

Second, however, the Court endorsed an intermediate filtering exercise before inventive step is assessed. Under G1/19, only features that contribute to the technical character of the invention, viewed as a whole, can support novelty or inventive step. Non-technical features that do not interact with technical subject matter in solving a technical problem do not provide a technical contribution.

The central question arising from today’s decision is therefore how far that intermediate stage can operate in practice before it becomes, in substance, another form of the contribution analysis rejected by the Supreme Court.

The UKIPO’s July 2026 examination guidance states that the Supreme Court’s judgment brought about an “immediate change“ in how section 1 of the Patents Act 1977 should be interpreted and applied, particularly for computer-implemented inventions. It identifies the Duns principles adopted by the Supreme Court as the basis for the new approach.

At the same time, the UKIPO’s Manual of Patent Practice continues to note that its detailed guidance is subject to change following Emotional Perception, with the relevant updates still pending as of July.

A new test, but a familiar outcome?

The immediate irony of the decision is that Emotional Perception has won a landmark Supreme Court ruling establishing a new legal framework for AI and computer-implemented inventions, only to have its own patent application refused again when that framework was applied on remittal.

That does not mean that the Supreme Court’s judgment has been undone. The UKIPO is now operating under a materially different legal test: the Aerotel approach has been rejected and the G1/19 framework, including its intermediate step, is now central to the assessment of computer-implemented inventions in the UK.

The dispute is instead likely to move to the next question: whether the UKIPO has applied that new framework in the manner intended by the Supreme Court and the EPO jurisprudence on which it relied.

For applicants developing AI, machine learning, and ANN technologies, that distinction could prove highly consequential. The initial eligibility hurdle may now be considerably easier to clear, while the treatment of technical and non-technical features at the inventive-step stage becomes the critical battleground.

The latest decision is therefore unlikely to be the final word on the practical consequences of Emotional Perception. Given the importance of the Supreme Court ruling and the concerns raised by the applicant’s representatives, further scrutiny — and potentially another appeal — could determine how closely UK practice ultimately tracks the EPO’s approach to AI and computer-implemented inventions.

← Back to News