UK Supreme Court shifts UK software/AI patent test: what it means for your business

On 11 February 2026, the UK Supreme Court handed down a landmark decision in Emotional Perception AI Ltd v Comptroller-General of Patents [2026] UKSC 3. The Court said that, when assessing whether a computer-related invention is excluded from patentability under Article 52(2) EPC (as applied in the UK), UK courts should move away from the long-standing “Aerotel” approach and instead align with the European Patent Office (EPO) approach often called “any hardware”.

That shift matters—especially for AI/machine learning and other software-heavy inventions—because it lowers the initial barrier to getting “through the door” on excluded subject matter. Whether a patent ultimately grants will still depend heavily on novelty and inventive step, but the route to that assessment is now more EPO-like.

1) What changed: Aerotel is no longer the starting point

The old position (in practice)

For years, the UK’s Aerotel framework was the main UK lens for “excluded subject matter” (e.g., computer programs “as such”). It was developed at a time when EPO case law was viewed as inconsistent, and the UK took a more distinct path.

The new position

The Supreme Court said UK courts should generally follow the EPO Enlarged Board of Appeal unless those decisions are clearly wrong or outside reasonable disagreement. And because the EPO has rejected the Aerotel-style reasoning in favour of “any hardware” logic (endorsed at Enlarged Board level, including in G 1/19), the UK should now use that approach too.

2) “Any hardware” in plain English

Under the “any hardware” approach, the question “is this excluded as a computer program?” becomes a very low threshold:

• If the claimed subject matter uses technical means (i.e., it runs on some form of computer/hardware), then it typically has technical character and is not excluded “as such” at the Article 52(2) stage.

This does not mean “software is always patentable.” It means software inventions are less likely to be thrown out early as excluded subject matter, and are more likely to be assessed on the usual patent tests: novelty and inventive step.

3) What the Court said about AI / neural networks (ANNs)

A central point in the decision is that an artificial neural network (ANN) is, legally speaking, treated as instructions that manipulate data to produce a result—so the Court accepted that an ANN is a “program for a computer.”

But crucially, the Court held that the computer program exclusion did not automatically apply because:

• under “any hardware,” the ANN must be implemented on computer hardware, and therefore the claim has the required technical character (so it is not a computer program “as such”).

Business takeaway: AI inventions—especially those implemented on conventional computing—should now have a clearer path to being examined on their technical merits rather than being blocked upfront by the exclusion framework.

4) Inventive step: the UK doesn’t fully copy the EPO, but it moves closer

A key debate in the case was: if the UK adopts “any hardware,” does it also have to adopt the EPO’s problem–solution approach for inventive step?

The Supreme Court’s answer was essentially:

• Not necessarily. The Court noted that EPC states don’t have to use the problem–solution approach, and the UK can keep its familiar Pozzoli/Windsurfing inventive step structure (for now).

However, the Court did say UK decision-makers should apply an EPO-style “intermediate step” before inventive step—i.e., identify which claim features actually contribute to technical character, and only then assess inventiveness using the UK approach.

And because the case had been argued for years under Aerotel, the Supreme Court sent aspects back to the relevant tribunal/decision-maker rather than fully applying that intermediate step itself on the facts.

5) What this means in practice for customers

Likely benefits

1. Fewer “excluded subject matter” knock-outs early on for software/AI claims that clearly run on hardware.
2. Closer UK–EPO alignment, which can simplify pan-European filing strategy and reduce “UK-only” surprises.
3. Better prospects for AI/ML claims to at least reach substantive examination on novelty/inventive step.

What doesn’t change

1. You still need a real inventive contribution – routine automation, generic ML training, or “use AI to do X” without a technical improvement may still fail on inventive step. (This is where many software cases succeed or fail.)
2. Drafting matters: the “technical character” filtering step means non-technical business/creative elements may be ignored when assessing inventiveness.

6) Practical guidance: how to position your AI/software invention now

If you’re looking to protect AI or software in the UK post-decision, the safer play is to frame the invention around technical contribution, for example:

• Improved computer performance (speed, memory, bandwidth, latency, energy use)
• Better training/inference efficiency (fewer parameters for the same accuracy, reduced compute, improved robustness)
• Better handling of real-world signals (sensor fusion, noise reduction, control outputs)
• System-level integration (how it operates in a deployed technical environment, not just a mathematical model)

And in claims/spec:

• Be explicit about how the ANN/software is implemented (pipeline, data structures, processing stages)
• Identify what is technically improved versus merely improved business outcomes
• Include fallback positions: method, system, computer-readable medium, and (where relevant) device-level claims

7) Quick FAQ

Does this mean “software is patentable in the UK now”?
Software has always been patentable in some circumstances, but this decision makes it less likely to be rejected early as excluded subject matter if it involves hardware implementation. Patentability will still hinge on novelty and inventive step.

Does this only help neural networks?
No. The reasoning applies broadly to computer-related inventions, with ANNs being a high-profile example.

Will the UKIPO immediately change examination practice?
The Supreme Court decision is binding law, but how it plays out operationally (especially around the “intermediate step”) may develop through UKIPO practice and further cases.