After Emotional Perception, Singapore Reopens the AI Patentability Question
IPOS has launched a feedback exercise on the UK Supreme Court’s Emotional Perception ruling, with submissions open from 17 April to 14 May 2026. The official questions go well beyond whether the UK case is merely interesting. IPOS is asking whether the decision is relevant to Singapore patent practice, whether it brings greater clarity to patentable subject matter and its interaction with inventive step, and what practical issues the approach could create for applicants and practitioners.
That is why this development deserves more attention than a routine consultation notice. Singapore is not changing its rules yet, but it is openly revisiting a pressure point that matters to AI companies: should ANN-related inventions continue to face heavy filtering at the subject-matter stage, or should more of the debate move toward technical character, technical effect and inventive step? For applicants working on accelerators, inference optimisation, edge deployment, signal processing and AI system architecture, that distinction can affect drafting strategy, fallback claim structure and prosecution cost.
Why this consultation matters more than it sounds
The wording of IPOS’s questions suggests that this is not a symbolic exercise. The second question is the real signal. It isolates the relationship between patentable subject matter and inventive step, which means IPOS is not only asking whether AI inventions should be easier to file. It is asking where the legal filter should sit. Should technicality be used aggressively at the threshold, or should more claims be allowed through that threshold and then tested on novelty and inventive step? The third question points to the administrative consequences: consistency of examination, disclosure burden, advisory cost and predictability for applicants.
Seen that way, the consultation is about examination design as much as policy. Singapore is effectively testing whether a more UK-and-Europe-style structure would produce better outcomes than a stricter front-end exclusion analysis for ANN-related inventions.
The UK ruling did not take ANNs out of software. It changed where the fight happens
A lot of commentary reduces Emotional Perception to a headline that “AI is now patentable”. That is too loose. The UK Supreme Court did not say that an ANN is outside the concept of a computer program. In fact, it held that an ANN is a program for a computer. The important shift was elsewhere: the Court rejected the long-used Aerotel approach and moved toward the EPO-style “any hardware” route. In practical terms, a claim that involves technical means is not excluded at the entrance simply because software is involved. It must still face the later questions of novelty and inventive step.
That changes prosecution dynamics. The core argument is less likely to be “is this software or not?” and more likely to become “what technical problem is being solved, what technical contribution does the claim make as a whole, and what evidence supports the alleged technical effect?” That is not a blanket liberalisation. But it does make room for some cases that previously died too early.
The likely beneficiaries are not all AI claims, but the ones tied to implementation
The EPO’s existing guidance is helpful here. AI and machine-learning models are still treated as abstract mathematical methods in themselves. But where a claim is directed to a method using technical means or to a device, the claim has technical character as a whole; and the algorithm can contribute to that technical character when it helps provide a technical solution to a technical problem. If Singapore leans in that direction, the main beneficiaries are unlikely to be generic claims about “using AI to improve results”. They are more likely to be applications tied to concrete implementation choices: ANN architectures for dedicated hardware or FPGA deployment, inference scheduling that improves latency or power use, classification of sensor or medical-monitoring signals, and training or deployment schemes written around real system constraints.
On the other hand, claims centred on recommendation logic, text organisation, business optimisation or abstract data grouping may still struggle if the technical use case is thin and the technical effect is asserted rather than shown. The line may move, but it will not disappear.
What applicants should fix now, before Singapore decides
The practical response is not to add the word “AI” more often. It is to improve the patent specification. Applicants should place the ANN inside a real technical system: identify the hardware environment, deployment location, data flow, resource constraints and measurable outputs. They should preserve layered claim positions as well, including method, system, device and medium claims where appropriate, so that a future shift in examination practice does not leave the application boxed into one formulation. And where the alleged technical effect depends on particular training-data characteristics, parameter ranges or implementation conditions, those enabling details should be disclosed early instead of being improvised later in prosecution.
This consultation may or may not lead to doctrinal change in Singapore. But it already tells applicants something important. The next contest in AI patenting is moving away from a simple entry question and toward a harder one: once the claim gets through the door, can the applicant show that it is genuinely a technical solution and not just an abstract model with a technical label attached?
