Software and artificial intelligence (AI) innovations are reshaping every sector of the economy. For businesses investing in these technologies, patents can be a critical commercial asset – they provide exclusive rights, deter competitors, and underpin licensing and investment strategies. If your business has invented something powered by artificial intelligence, AI, such as a medical diagnostic tool that uses image recognition to analyse images, or a cybersecurity tool that uses a large language model to check for threats, you may want to know whether you can patent it, and where. Yet the question of whether software or AI can be patented remains one of the most misunderstood areas of intellectual property law.
The short answer is yes. It is generally possible to protect software and AI in all the major patent-granting jurisdictions. However, the requirements differ.
This article provides an overview of how AI and software can be patented in key jurisdictions and commercial markets. This will help you and your trusted patent attorney to develop an optimal patent filing strategy for your AI innovations.
Introduction
Why are software and AI inventions treated differently to, say, mechanical inventions?
Firstly, there’s no such thing as a ‘worldwide patent’ – this means, patents are granted by each patent-granting jurisdiction separately, according to their own patent laws. Each patent-granting jurisdiction has the same basic rules about what is allowed to be patented. An explanation of these ‘patentability’ rules can be found in this glossary. The first of these is whether an invention is allowed to be patented at all under the local law. The law in many jurisdictions singles out “computer programs” and “mathematical methods” as things that are not allowed to be patented as such, which impacts AI-based innovations. This is why, compared to mechanical or chemical inventions, the first hurdle to overcome is to determine whether a specific software or AI invention is allowed to be patented.
With this in mind, this article presents how nine key jurisdictions view software or AI inventions.
1. United Kingdom
A. Summary
Software and AI innovations can generally be patented in the UK, and a landmark 2026 Supreme Court decision has made it significantly more straightforward to do so.
B. What the patent law says
Under the UK Patents Act 1977, a patent cannot be granted for “a program for a computer” as such. The phrase “as such” has always been important. It signals that what is excluded is a computer program that makes no technical contribution. For years, the difficult question was how to determine if a program makes a technical contribution. This question was answered by various court decisions, which are summarised below.
The Former Approach: Aerotel (2006)
From 2006, UK Intellectual Property Office applied the four-step “Aerotel test” to assess whether a software or AI invention made a technical contribution or should be excluded from patentability. This was often quite difficult to satisfy. Consider an AI model that provides you with recipe recommendations based on previous recipes you tried, or restaurants you visited and liked. This would likely be considered not technical because the model makes a human’s life easier, and does not necessarily itself provide a technical advantage.
The Current Approach: Emotional Perception AI (2026)
In February 2026, the UK Supreme Court stated that the Aerotel test is to be abandoned in favour of a new approach. The new approach, which has not yet been fully worked-out, is intended to align the UK approach to software and AI inventions with the European Patent Office (EPO), and in particular, EPO’s decision in the case G1/19 (see below). As the Supreme Court has left it up to the UK IPO to determine how to align their analysis of software and AI inventions, it will likely be some months before we see how this works in practice. However, in the meantime, European patent attorneys are well-versed in how the EPO analyses such inventions and can advise you on your chances of success in both jurisdictions.
What this means in practice is that a wider range of software and AI inventions will now be properly searched and examined by the UK IPO, rather than being dismissed immediately as being excluded from patentability.
Thus, certain AI inventions with a genuine technical effect that previously may have been dismissed (such as new neural network architectures, novel training methods for machine learning models, or recommendation systems), can now be assessed in terms of their novelty and inventiveness. This does not mean they will be granted as patents, but they will at least overcome the initial barrier in the process.
2. European Patent Office (EPO)
A. Summary
The European Patent Office makes it possible to obtain patent protection in 39 member states via a single application, which is very useful if you intend to commercialise your invention in multiple states. Software and AI innovations can generally be patented at the EPO as long as they provide a technical solution to a technical problem.
B. What the patent law says
Just like in the UK, under the European Patent Convention 2000, a patent cannot be granted for “a program for a computer” as such. The EPO has also attempted to answer the difficult question of how to determine if a program makes a technical contribution. The EPO’s approach is summarised below.
The EPO has a two-stage approach to assessing software and AI inventions, which was reiterated in G1/19:
Firstly, the EPO considers whether a software or AI invention involves any hardware, such as a computer, a processor, a memory. If so, the EPO takes the view that the invention is not merely a computer program as such, and should proceed to be assessed for novelty and inventiveness. This is why we usually start patent claims for software or AI inventions with the words “A computer-implemented method for…”, because it ensures the claim satisfies this hurdle at the EPO.
Secondly, the EPO considers whether the software or AI invention is inventive in view of the prior art (see definition of “prior art” here). Here the EPO’s approach from the COMVIK case applies: features of the claim that are non-technical (e.g. business methods, presentation of information, abstract mathematical steps, etc.) are not counted when assessing inventive step. The remaining features (if any) must make a technical contribution over the prior art in order to demonstrate that the invention is inventive. Examples of technical contributions include controlling a machine, improving computer performance, reducing processor or memory usage, enhancing data security, improving communication efficiency, and so on.
Consider, for example, a machine learning method that is built to quickly classify financial transactions in real-time, in order to detect fraudulent transactions fast. This is not only beneficial for banks but also their customers. Business methods are not considered technical, but focussing the patent claims on how the machine learning model is trained and deployed (i.e. the steps involved, the underlying architecture, etc.) to very quickly analyse thousands of real-time transactions to identify possible fraud or malicious activity, may make it easier to demonstrate the technical contribution of the invention.
It remains difficult to patent certain inventions involving natural language processing at the EPO. Often, this is because the EPO considers certain natural language tasks, such as translation from English to German, summarisation, or even classification of documents, to be based on linguistics and natural language rather than something technical. Of course, some highly complex models are being used for natural language processing tasks, including large language models, but the EPO’s stance remains the same. Consequently, it can be very hard to identify a technical contribution when the output of a method is just a string of words.
3. United States of America
A. Summary
Software and AI innovations can generally be patented in the USA, and a 2025 case has broadened the types of AI inventions that are patent-eligible.
B. What the patent law says
Under 35 USC 101, patents may be granted for “any new and useful process, machine, manufacture, or composition of matter”. The US Supreme Court has interpreted this as meaning that abstract ideas, laws of nature, and natural phenomena are not allowed to be patented (i.e. are “judicial exceptions”).
In particular, in 2014 a US Supreme Court decision (Alice Corp v CLS Bank International) introduced a framework for assessing whether software inventions are permitted to be patented or are judicial exceptions. The Alice case, and how the framework was differently applied in practice, led to considerably uncertainty in whether software could be patented.
That landscape shifted significantly in late 2025. The Desjardins case is a landmark precedential decision which states that inventions that improve the functioning of machine learning models themselves should not be dismissed as abstract ideas. This is because there are technical advantages, such as reduced storage, lower model complexity, improved multi-task performance. This case suggests that improvements in the functionality of algorithms and computer programs themselves make an otherwise abstract idea something that is integrated into a practical application. This means that software or AI inventions that lead to a technological improvement can generally be patented as long as the claims recite that technological improvement or the features that cause that improvement. It is also important that the technical problem and technical solution provided by the invention is clearly described in a patent application, as this makes it easier to demonstrate a software or AI invention is not abstract.
4. China
A. Summary
Software and AI innovations can generally be patented in China, though since January 2026, some types of invention are not allowed to be patented on ethical grounds.
B. What the patent law says
China’s patent law states that a patentable invention is “a new technical solution relating to a product, a process, or an improvement thereof”.
With respect to software and AI inventions, there is a three-part check to determine whether the inventions provide a technical solution. The first part asks whether the invention addresses a technical problem. The second part asks what technical approach does the invention use to address the problem. Finally, the third part asks what measurably better outcome is obtained as a result. Thus, China’s approach has considerable overlap with the EPO’s approach.
In 2024, new patent examination guidelines were published in relation to AI inventions, and these were updated in January 2026. These updated guidelines state that AI innovations that violate laws or social ethics are not allowed to be patented. Examples of innovations that would not be awarded a patent under these guidelines include a facial recognition system for targeted commercial marketing in shopping malls, and an autonomous vehicle emergency system that factors in a pedestrian’s age and gender when deciding how to respond.
In addition, the updated guidelines specify that an innovation that simply applies known AI techniques to new areas without making any technical modifications to those known AI techniques, is unlikely to be inventive. Again, this is quite similar to the EPO’s approach to assessing inventive step.
5. South Korea
A. Summary
Software and AI innovations can generally be patented in South Korea.
B. What the patent law says
South Korea patent law generally requires that software and AI inventions can be patented if they are implemented using hardware.
There are generally considered to be three types of AI invention. The first type is for core AI. This type comprises inventions where the novel contribution is the model architecture or the training methodology itself. The second type is for applied AI. This type comprises inventions where AI is used to solve a problem such as autonomous driving, medical healthcare, etc. The third type is for inventions by AI, meaning inventions mostly created by AI itself. These inventions cannot be protected because in South Korea (and in most major jurisdictions), it is required to list only human inventors when applying for patents.
6. India
A. Summary
Software and AI innovations can generally be patented in India.
B. What the patent law says
Indian patent law has overlaps with UK patent law, in terms of what is excluded from patentability. Computer programs and algorithms as such are not allowed to be patented. This exclusion has historically been interpreted broadly, which means that many AI-related inventions have prima facie been caught under the exclusion.
In July 2025, the Indian Patent Office published new guidelines specifically including AI related inventions for the first time. These guidelines provide worked examples and explain that AI inventions are more likely to be acceptable if they provide technical advances such as boosting internal system efficiency, or are linked to concrete technical implementations. Similarly, the guidelines mention that AI inventions are more likely to be excluded if they relate to presenting data visually or textually, mimicking human reasoning or decision making, or recommendation systems based on user preferences.
7. Australia
A. Summary
Software and AI innovations can generally be patented in Australia, if implemented on hardware.
B. What the patent law says
Australian patent law states that a patent must be for “a manner of manufacture within the meaning of section 6 of the Statute of Monopolies 1623”. This means that what is considered patentable in Australia is based on extremely old law, and the courts have had to determine how to interpret the law to deal with much more modern technologies such as software.
In 2025, the Australian Full Court decision in the long-running dispute between Aristocrat Technologies and the Commissioner of Patents (“Aristocrat”) set out the test for patentability for software inventions. The court stated that a software invention can be patented if it is manipulated on a computer or is implemented on a computer to produce a useful result. In February 2026, the High Court refused to allow the Commissioner of Patents to appeal this decision. This suggests that as long as software or AI inventions involve some hardware, they are likely to satisfy the patentability requirement. This is not unlike the EPO’s approach.
8. Japan
A. Summary
Software and AI innovations can generally be patented in Japan, if implemented on hardware.
B. What the patent law says
Japanese patent law defines an invention as “a highly advanced creation of technical ideas utilising the laws of nature”. For software and AI, this usually means that the invention has to make specific and concrete use of hardware, such as processors and memory.
The Japanese Patent Office, JPO, published detailed case examples specifically for AI related technologies (including a Manga version of the guidelines!). For example, training data used in machine learning, by itself, constitutes merely the presentation of information and does not satisfy the requirement for patentability. However, a model that has been trained, using that training data, to generate images, and which is concretely implemented using hardware resources, is more likely to meet the patentability requirement.
9. Canada
A. Summary
Software and AI innovations can generally be patented in Canada.
B. What the patent law says
In Canada, following the recent decisions in the “Benjamin Moore” cases, software and AI inventions can be patented as long as the invention improves the functioning of the computer or as long as the invention involves non-generic computer hardware. This can be a high bar to satisfy, and is similar to the situation in the UK until this year. Inventions that decrease memory requirements, increase processing speed, reduce latency, and so may be more likely to satisfy the patentability requirement in Canada.
The Canadian Patent Office has provided guidance on AI including example AI inventions that may satisfy patentability. For example, a machine learning model that processes weather data and outputs recommendations on when to irrigate is not patentable because it is considered abstract. However, if the invention actually causes an irrigation system to turn on or off in response to the recommendations, this is more likely to be patentable.
Conclusion and key take aways
Software and AI inventions are generally patentable in many of the key patent granting jurisdictions in the world. However, in all jurisdictions it is necessary to demonstrate that the software or AI invention provides a technical solution to a technical problem. It is also clear that the law is evolving everywhere as law makers grapple with how to encourage innovation and how to update or interpret existing patent laws for modern technologies.
Practically, European patent attorneys are well placed to craft patent applications for software and AI inventions, because the “technical problem-solution” approach is well-established law in Europe. If you are innovating in software or AI and want to understand what protection is available for your technology, our AI team would be pleased to help. We can devise a patent filing strategy to suit your business and the differing global requirements on patentability.
Souce:
https://www.appleyardlees.com/software-and-ai-patentability-a-global-overview/?utm_campaign=Articles&utm_medium=email&_hsenc=p2ANqtz-8OKDMtiNnI-9ZaXDvRKuBsc2PPUgLqPzJeV_5OYhiUR3EY2jfYEQfGztssAecD54Gt-y7W563ikpUG3X8_V9SSsIe5nQ&_hsmi=411739343&utm_content=411739343&utm_source=hs_email