Artificial intelligence is changing the way businesses operate — from automating customer service to generating entirely new products and processes. But here is a question that is becoming increasingly urgent: if an AI system creates something new, can you patent it? And if you build a business around an AI-driven innovation, are you actually protected?
The answer, at least under current UK law, is nuanced. Recent court decisions have been grappling with exactly these questions, and the outcomes matter enormously for business owners and entrepreneurs who are investing in AI-based innovation. This article breaks down the key cases and what they mean for you in plain English.
First, the Basics: What Is a Patent and Why Does It Matter?
A patent gives you the exclusive right to prevent others from making, using, or selling your invention for up to 20 years. For businesses, patents are a powerful commercial weapon — they let you protect market share, license your technology to others for revenue, and deter competitors from copying you.
In the UK, patents are governed by the Patents Act 1977 and administered by the Intellectual Property Office (IPO). But not everything can be patented. Among the things explicitly excluded from patent protection are computer programs and mental acts — and it is this exclusion that has caused enormous difficulty in the context of AI and software-based inventions.
“The critical question the courts keep asking is this: does your AI-driven invention make a real-world technical contribution— or is it just software doing what software does?”
The Aerotel Framework: The Four-Stage Test Every AI Patent Must Pass
The starting point for understanding how UK courts assess AI-related patents is the case of Aerotel Ltd v Telco Holdings Ltd [2007] RPC 7. This case—decided by the Court of Appeal—established a structured, four-stage test that patent examiners and courts must work through whenever an application involves potentially excluded subject matter, such as software or computer programs.
To put it simply, the four stages are:
• Properly understand what the patent claim is actually saying.
• Identify what the real-world contribution of the invention is — what has it actually added to the world?
• Ask whether that contribution falls entirely within the excluded categories (like being a pure computer program or mental act).
• Even if it seems to fall within exclusions, determine whether it nonetheless makes a genuine technical contribution.
For business owners, the key takeaway from Aerotel is this: the mere fact that your innovation runs on software or uses AI does not automatically exclude it from patent protection. What matters is whether there is a real, technical effect in the world beyond the software itself. If your AI system controls a physical manufacturing process more efficiently or analyses medical images in a way that produces better clinical outcomes, that technical effect could be patentable.
The problem arises when the innovation is essentially the software logic itself—a clever algorithm, a new way of organising data, or a novel AI model—without any tangible technical effects outside the computer. That, as courts have consistently held, is not patentable under UK law.
The AT&T Signposts: Five Questions to Ask About Your AI Invention
The case of AT&T Knowledge Ventures v Comptroller of Patents [2009] FSR 19 further refined the approach established in Aerotel. Mr. Justice Lewison identified five “signposts”—practical indicators to help determine whether an invention makes a genuine technical contribution. These signposts have become an important part of how both the IPO and the courts assess AI and software patent applications.
In practical terms, the five signposts ask the following:
• Does the invention produce a technical effect that goes beyond the normal physical interactions of running a program?
• Does it produce a technical effect on a process outside the computer — for instance, controlling machinery or improving a communications network?
• Does it solve a technical problem within the computer itself — such as running more efficiently or processing data faster?
• Is the contribution made by a new form of programme rather than just a new programme doing the same old things?
• Would the contribution be patentable if it were implemented in hardware rather than software?
For entrepreneurs, this last signpost is particularly useful as a mental test. Strip away the software and ask, if this innovation were built into a physical machine or circuit, would it be patentable? If the answer is yes, there is a strong argument that your AI-driven version deserves the same protection.
Tip for Business Owners: Before investing in a patent application for an AI or software-based innovation, ask your IP solicitor to map your invention against these five signposts. This will give you a realistic picture of your prospects before you spend money on the application process.
The Game-Changing Case: Emotional Perception AI Ltd v Comptroller [2023]
The most significant and recent development in this area came in the case of Emotional Perception AI Ltd v Comptroller-General of Patents [2023] EWHC 2948 (Ch), decided by Mr Justice Mann in the High Court. This case has profoundly impacted the IP world and significantly alters the landscape for businesses utilising AI, especially those employing machine learning and neural networks.
Emotional Perception AI had developed a system that used an artificial neural network (ANN) to recommend music files to users based on semantic similarity – essentially, recommending songs that ‘feel’ similar to one another rather than simply matching genre or artist tags. The IPO refused the patent on the basis that it was a computer program as such and, therefore, excluded from protection.
Mr Justice Mann disagreed. His reasoning is fascinating and worth understanding, because it fundamentally challenges the way we classify AI systems under patent law.
“The judge held that a trained neural network is not simply a computer program in the conventional sense. Once trained, it operates more like a bespoke machine — and the rules the network applies are not written by a human programmer but emerge from the training process itself.”
This distinction matters enormously. Traditional software exclusions were designed with conventional computer programs in mind—code written by humans and following explicit logical instructions. A neural network, the Judge reasoned, is different in nature. After training, the network’s internal ‘rules’ are not human-authored; they emerge from patterns in data. Treating such a system as simply a ‘computer program as such’ may therefore be wrong in principle.
The judge also discovered that the system generated a technical effect, leading to the recommendation and transmission of specific files. Even if the ultimate user experience was the emotional or semantic quality of the recommendation, the mechanism producing it was technical in character.
What This Means for Your Business
If you are building a product or service that relies on machine learning, neural networks, or other AI technologies, the Emotional Perception case is potentially excellent news. It suggests that AI-driven innovations may have a stronger claim to patent protection than previously thought — at least where the AI system itself is novel and produces a real-world output.
However, there are important caveats. The case is a first-instance High Court decision, and the IPO has indicated that it will continue to apply the Aerotel framework while the broader legal position is clarified. The possibility of an appeal or the emergence of further guidance is highly likely. The law in this area is genuinely in flux.
Here is what we recommend for business owners right now:
• Do not assume your AI innovation is automatically unpatentable just because it involves software or machine learning. The law has moved and may move further.
• Do not assume it is automatically patentable either. The technical contribution question remains central, and weak applications will still fail.
• Document your innovation carefully — including how the AI system was trained, what data was used, and what technical outputs it produces. This documentation will be critical if you pursue a patent application.
• Consider a freedom-to-operate analysis alongside any patent strategy. Even if you can’t patent every part of your AI system, it’s vital to know what others have patented.
• Seek specialist advice early. Patent strategy for AI is still evolving rapidly, and generic advice from non-specialists can be costly.
What About AI as an Inventor? The DABUS Problem
There is one further dimension worth mentioning. All of the cases discussed above involve human inventors using AI as a tool. A separate and equally fascinating legal question is whether an AI system can itself be named as the inventor of a patent.
The UK Supreme Court addressed this in Thaler v Comptroller-General of Patents [2023], confirming that under current UK law, an inventor must be a human being. A patent application that names only an AI as the inventor will be denied. This ruling has significant implications for businesses where AI plays a central generative role in the innovation process — the humans overseeing, directing, and deploying the AI system need to be identifiable and nameable as inventors.
This issue is an area where the law may yet evolve and where international differences are already emerging. Businesses with global IP strategies need to keep a close eye on developments in the US, EU, and other key jurisdictions.
info@lawdit.co.uk
