Background
Emotional Perception AI Ltd filed a UK patent application for a system that uses an artificial neural network (ANN) to recommend media files — such as music or video — based on their emotional characteristics. The system stores media files in a database, uses a trained ANN to determine emotional associations between files, and then recommends content to users via a communications network based on how emotionally similar it is to media they have previously enjoyed.
The UK Intellectual Property Office refused the application on the ground that it related to a “program for a computer… as such” — one of the categories excluded from patentability under section 1(2) of the Patents Act 1977. The High Court reversed the refusal, but the Court of Appeal reinstated it, applying the long-established Aerotel/Macrossan four-step test (2006) and concluding that the “actual contribution” of the invention was merely a computer program performing calculations.
Emotional Perception AI appealed to the Supreme Court, arguing that the Aerotel framework was out of step with how the European Patent Office (EPO) assesses similar inventions.
The Court’s Holding
The Supreme Court unanimously allowed the appeal, holding that Emotional Perception’s ANN-based system is patentable subject matter.
In a landmark departure from two decades of UK case law, the Court abandoned the Aerotel/Macrossan four-step test and adopted the EPO’s approach established in COMVIK and endorsed in enlarged board decision G1/19. Under the new framework:
- Step 1 (“Any Hardware” approach): Claims that reference hardware — even method claims specifying execution on a computer system — qualify as “inventions” and are not excluded as programs for a computer “as such.” The Court held that the ANN claims, which recited a database, a communications network, and a user device, cleared this threshold.
- Step 2 (Intermediate assessment): The patent office must identify which features “contribute to the technical character of the invention,” viewing the invention holistically and recognizing that non-technical features may contribute when they interact with technical elements.
- Step 3 (Robustness check): Novelty, inventive step, and industrial applicability are assessed based on the features identified as contributing to technical character.
The Court concluded that the old Aerotel approach — which asked whether the “actual contribution” of the invention was technical — had diverged from EPO practice and created artificial barriers to computer-implemented invention patents in the UK that did not exist in other major jurisdictions.
Key Takeaways
- The UK now applies essentially the same framework as the EPO for assessing computer-implemented inventions — a “permissive initial eligibility threshold” paired with rigorous scrutiny at the inventive step stage.
- Patent applicants in the UK should recite specific hardware elements (databases, networks, processors, user devices) in their claims. Even generic hardware references should now suffice to clear the eligibility bar.
- The UK IPO has announced it will reopen and reassess previously refused applications under the new standard, potentially unlocking a wave of AI and software patent grants.
- The decision parallels developments in US patent law (the USPTO’s Ex parte Desjardins framework from December 2025), as both jurisdictions move toward permissive initial eligibility with robust downstream scrutiny.
Why It Matters
This is the most consequential change to UK patent law for computer-implemented and AI inventions in twenty years. The Aerotel test had made the UK one of the hardest jurisdictions in which to patent software and AI innovations — applicants routinely obtained European patents that their UK national counterparts refused. By aligning with the EPO’s COMVIK approach, the Supreme Court has removed a significant barrier to AI patent protection in the UK, just as the country seeks to position itself as a hub for AI innovation post-Brexit. For AI companies choosing where to file, the UK is now back in play as a viable patent jurisdiction for neural network inventions, machine learning systems, and other computer-implemented innovations that would previously have been rejected as unpatentable programs “as such.”