The UK Intellectual Property Office has refused a patent application that the Supreme Court had sent back to it. Decision BL O/0894/26 of 1 October 2026 concerns application GB1904713.3 in the name of Emotional Perception AI Ltd for a computer-implemented artificial neural network that recommends files, mainly music, matched by meaning. The Supreme Court had held that the network was a computer program but was not excluded by section 1(2) of the Patents Act 1977 because it included computer hardware, and it remitted the application to the Office to work through the remaining steps of the European Patent Office approach in Duns, G1/19. The refusal is a first instance disposal by a hearing officer and now stands as the outcome of the application.
Hearing officer Mr S Brown took the claim through sections 1(1) and 1(2) of the Patents Act 1977 and then the inventive step enquiry, applying the Duns approach the Supreme Court had identified. He found that producing a semantically better file is not a technical result. Because the training details of the network produced no technical effect, they could add nothing to the technical character of the claim, and they were disregarded for the purposes of inventive step. The features that remained were found obvious over common general knowledge. The application was refused under section 18(3) of the Patents Act 1977.
Developers filing in the United Kingdom on recommendation engines, ranking systems and model training methods face a narrower path than the Supreme Court ruling on the exclusion point appeared to open. Getting past section 1(2) by claiming a trained network running on hardware leaves the inventive step enquiry untouched, and training methodology counts there only where it delivers a technical effect beyond the quality of the model's own output. Patent attorneys drafting for machine learning clients will need claims that locate a technical effect other than improved relevance or semantic matching. Licensors, acquirers and investors valuing United Kingdom AI patent families should treat claims resting on output quality alone as exposed.
The decision leaves open what counts as a technical result where a model's measured gain shows up only in the quality of what it returns. The refusal rests on inventive step and not on the computer program exclusion, which the Supreme Court had already resolved in the applicant's favour on the hardware point, so the patentability of AI implemented inventions as a class is not disturbed. The published decision records the ground of refusal as section 18(3) of the Patents Act 1977, lists the provisions discussed as sections 1(1) and 1(2), and carries the keywords excluded fields and inventive step.
Licentium advises AI developers, patent attorneys and investors on protection and enforcement of machine learning inventions in the United Kingdom and comparable jurisdictions. Work we undertake includes reviewing claim sets for technical effect beyond output quality, assessing United Kingdom and European Patent Office prospects for model training and inference claims, preparing responses to examination and hearing objections, and diligence on AI patent families in licensing and acquisition.