Navigating AI Patents: key updates from EPO and UKIPO Guidelines

Both the EPO and the UKIPO included some specific updates for AI patenting in their Guidelines for 2025. The UKIPO added some guidance to its examiners for quantum computing. These 2025 changes encourage us to recommend using the EPO rather than the UKIPO when seeking UK patent protection for the majority of AI inventions and inventions based on quantum computing algorithms, at this point in time, and yet it is sometimes possible to obtain broader claims via the UKIPO due to the EPO’s strict assessment of which features are “essential”. We will let our clients know if and when this changes. We will continue monitoring case law developments (including the case of Emotional Perception AI Ltd that is due to be reviewed by the UK Supreme Court in 2025) and we will continue working with the EPO and national patent offices to encourage increased harmonisation and consistency. We believe consistency and predictability benefits all industries.

EPO Guidelines

In recent years, the EPO’s Guidelines for Examination have become an increasingly useful tool for Examiners and applicants, because the level of detail of the guidance to Examiners has increased and because the EPO has implemented an annual review cycle which opens the Guidelines to public scrutiny and reflects recent Board of Appeal case law.

In 2025, the EPO has improved its Guidelines for examination of AI cases, with an update to section G-II, 3.3.1.

The European Patent Office Guidelines on patentability of mathematical methods benefitted from a substantial rewrite in 2018, after the EPO reviewed around 1000 Board of Appeal decisions and distilled the patentability guidance of the Boards to give us very helpful guidance in Section G-II, 3.3. This was clarified in early 2022, following the EPO Enlarged Board of Appeal’s reasoned decision G1/19 in 2021 – a computer simulation case led by Venner Shipley LLP. This case told us that the EPO’s well-established approach to the assessment of computer-implemented inventions was equally applicable to computer simulation cases, and identified different ways in which a claimed invention can satisfy the requirement for a technical effect. The Guidelines explain that, although a purely abstract mathematical method which does not involve technical system features is not patentable, a claim to a method involving use of a computer or a claim to a device is not excluded from patentability, and any features which contribute to the technical character of an invention and the achievement of technical effects can also contribute to its inventive step. In particular:

“A mathematical method may contribute to an invention’s technical character, i.e. contribute to producing a technical effect that serves a technical purpose, by being applied in a field of technology and/or by being adapted to a specific technical implementation”

To qualify as a “specific technical application”, the claims must be functionally limited to the particular application and technical purpose so as to achieve a technical effect. The EPO Guidelines provide helpful examples such as control of an apparatus or industrial process, enhancing digital audio/image/video, separating/recognizing speech, or encoding data for reliable transmission such as error-correction coding or compression or encryption, etc). A “specific technical implementation” is where a mathematical method is adapted to a specific implementation in the sense that its design is motivated by the technical features and internal functioning of the system or network on which it is implemented – to exploit system capabilities or address system constraints. Again, useful examples are given. This has not changed.

The change comes in the next section of the EPO’s Guidelines, section G-II, 3.3.1 which relates specifically to artificial intelligence and machine learning. This has now also been updated with wording that is much more consistent with the principles of the previous section. This is helpful because it removes an inconsistency and an unintentionally negative impression about patentability of AI inventions. The new wording is not exactly what we recommended, but we are grateful to the EPO for listening to our recommendations and increasing consistency.

Section 3.3.1 of the EPO’s Guidelines for Examination now includes these comments:

• with reference to computational models and algorithms which are stated to be of an abstract mathematical nature: “their use does not by itself render inventions related to artificial intelligence or machine learning nonpatentable”; and
• “if a claim of an invention related to artificial intelligence or machine learning is directed either to a method involving the use of technical means (e.g. a computer) or to a device, its subject-matter has technical character as a whole and is thus not excluded from patentability under Art. 52(2) or (3). In such cases, the computational models and algorithms themselves contribute to the technical character of the invention if they contribute to a technical solution to a technical problem, for example by being applied in a field of technology and/or by being adapted to a specific technical implementation”.

We do not anticipate the new wording leading to a major change of EPO practice for Examining Divisions that are already familiar with AI cases, because this wording mainly reflects the established practice set out in previous section G-II, 3.3. Nevertheless, the change is a positive one and will be helpful for Examiners who are less familiar with examination of AI cases, some of whom appeared to have been influenced by the previous inconsistent and overly negative wording of section 3.3.1.

The EPO also made a small improvement to section F-III, 3 of the Guidelines, deleting some very negative case-specific comments from Board of Appeal case T1191/19 about the insufficiency of the description of that computer-implemented meta-learning case. Regardless of whether the Board of Appeal’s comments were valid for that specific case, we suggested that they were not appropriate as a statement of general principles in the Guidelines. Nevertheless, the Board of Appeal’s negative comments in cases T161/18 and T1191/19 serve as an important warning to everyone drafting AI-related cases, and the 1 April 2025 Guidelines retain these comments:

“Another example [of insufficiency] can be found in the field of artificial intelligence if the mathematical methods and training datasets are disclosed in insufficient detail for the skilled person to be able to reproduce the technical effect without undue burden using common general knowledge over the whole scope of the claim (see G-II, 3.3.1).”

UKIPO Guidance

On 2 January 2025, the UK Intellectual Property Office (UKIPO) updated its Manual of Patent Practice (MoPP). On 30 January 2025, the UKIPO published an update to its guidelines for examining patent applications relating to AI. The updates are available on the UKIPO website:

The UKIPO should be praised for producing such a clear and detailed MoPP, with references to relevant case law. The UKIPO’s MoPP and AI patenting guidance set out how they will apply the guidance they have received from the UK courts, with clear explanations of the case law they are relying on and numerous example scenarios. This is helpful for examiners and applicants. However, that does not mean that we are enthusiastic about UKIPO practice in this area.

The guidance on computer programs has been updated to take into account the decision by the Court of Appeal in the case of Comptroller-General of Patents, Designs and Trade Marks v Emotional Perception AI Ltd [2024] EWCA Civ 825, and thus reflects a more negative practice for AI inventions compared to the permissive approach adopted in the earlier (overturned) High Court decision.

The MoPP section 1.35 now references the very wide definitions of ‘computer’ and ‘program for a computer’ set out in the Court of Appeal judgement. According to these definitions, a computer is “a machine which processes information”, and a program for a computer is “a set of instructions for a computer to do something”.

In the Emotional Perception judgement, an artificial neural network (ANN) was considered to fall within this definition of a computer, and the weights applied by training the ANN were found to be nothing more than a computer program. Technical implementation details were largely disregarded once the overall purpose was identified and considered not to provide a technical contribution.

Section 1.35.2 of the MoPP notes that ‘hard-wired’ computers such as application-specific integrated circuits (ASICS) are also considered to be computers, such that instructions stored on such circuits will be assessed in the same way as other computer programs, and section 1.35.3 notes that neural networks, analogue computers and quantum computers are all also within the broad definition of ‘computer’ set out by the Court of Appeal. The conclusion is that programs for controlling them should be assessed in the same way.

Section 1.35.5 notes the importance of determining whether a claimed computer program provides a technical contribution. Section 1.35.6 notes that claiming conventional computing hardware or claiming a program on a carrier is not sufficient to confirm the presence of a technical contribution to the art. Sections 1.36 to 1.45 remain helpful as they summarize the role of the “AT&T signposts”[1] in assessing technical contribution.

The guidance remains consistent with an earlier statutory guidance note which recognised the possibility that technical contribution could potentially arise in the case of a “better ANN”, where the invention is said to improve the functioning of the ANN itself (while at the same time emphasising that the signposts should be applied carefully in such cases). This possibility, in which an AI invention itself is considered to be “better computer”, is explored in more detail here.

For the first time, the UKIPO has added sections specific to quantum computing, the first two sections of which are copied below:

“1.46

As noted in 1.35.3, the definition of a computer and computer program cover quantum computers and their programs. As a consequence, applications involving programs for quantum computers will be assessed by applying the Aerotel steps along with the guidance of the signposts in AT&T and HTC v Apple.

Quantum computers are analogous to their classical counterparts in that they can be programmable. They can be adapted to implement specific hard computations like classical co-processors (for example in classical-quantum hybrid techniques such as Variational Quantum Eigensolver, VQE). Given these analogies with the classical scenario, the signposts are useful in the quest to find a technical effect or contribution for quantum computer programs.

1.46.1

Following the guidance of the signposts, there are generally no issues of patentability under section 1(2) when a computer program, either quantum or classical, makes a contribution towards a quantum computer operating in a new or improved way. Equally, there are no issues when the computer program results in a technical effect at the architectural level of the quantum computer. In summary, when a computer program is about the physical implementation and/or operation of the hardware, e.g. aspects of superconducting and/or photonic circuits or gates for controlling, processing and measuring qubits it is likely to escape exclusion.

By way of example, programs for quantum computers may escape exclusion when: they are applied to a technical process, i.e. if the program results in a technical effect in a non-excluded area; or the program achieves a technical effect at the core or implementation level of the quantum computer or quantum/classical hybrid systems.

A relevant technical effect may also be found in detecting and correcting errors of the quantum computer. Mitigating a shortcoming of quantum hardware in the current Noisy Intermediate-Scale Quantum (NISQ) era by adapting the program design, e.g. at the logical qubit level and/or by compiling or transpiling, may also result in a relevant technical effect if it does so by overcoming rather than merely circumventing the perceived technical problem (as per signpost v)), e.g. where the problem is short decoherence time and/or noise mitigation.”

This is followed by references to a small number of UKIPO Hearing Office decisions for quantum computing cases which were determined not to make a technical contribution, but which noted that a technical application involving the use of a quantum computer may have a relevant technical effect, if the use is clearly described and claimed. The Hearing Officer in BL O/1193/23 (which related to identifying a valid excited state of a system of interacting electrons using a hybrid quantum-classical system) concluded that the normal operations of a quantum computer that might not be excluded as a “computer program” (i.e. executing a method involving changing the state of physical qubits, and determining the states by measurement) were not enough for patentability – the contribution of the claimed invention was considered to be an algorithm implemented in software.

If you have specific questions relating to patentability of inventions, please contact us.

[1] From the case of AT&T Knowledge Ventures/Cvon Innovations v Comptroller General of Patents [2009] EWHC 343 (Pat), as restated in HTC Europe Co Ltd v Apple Inc (Rev 1) [2013] EWCA Civ 451.