Andrew Stutesman, Unsplash
Artificial intelligence may be global, but patent eligibility remains stubbornly local. A recent decision out of the Supreme Court of the United Kingdom seems to have nudged UK practice for computer-implemented inventions closer to the approach historically taken by the European Patent Office. The decision lowers the threshold for exclusion from patentability, reducing the likelihood that applications for these types of inventions will be automatically rejected on their face. However, this approach still differs from U.S. practice, representing a continental divide that needs to be taken into account when seeking patent protection for AI-related inventions on a global basis. AI prosecution should not occur in cross-border silos.
In Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Marks, the UK Intellectual Property Office rejected a patent application seeking protection for an artificial neural network trained to align physical properties of media files with human-perceived emotional responses in order to improve AI-driven recommendation engines. The UKIPO held that “schemes, rules and methods for performing mental acts, playing games or doing business, and programs for computers” are simply not regarded as patentable inventions. The case progressed through the England and Wales High Court of Justice and the Court of Appeal (England and Wales) before the Supreme Court issued its decision on February 11, creating a new standard for patent eligibility in the UK based on EPO standards.
For years prior to this decision, applicants in the UK have faced a familiar frustration. If an invention even smelled like software, it risked being categorized as a “computer program as such” and excluded from patentability before novelty or inventive step received serious attention. In Emotional Perception, the court rejected the idea that an artificial neural network should be treated as a computer program for exclusion purposes merely because it is implemented in software on a standard computer. Put differently, the court was willing to look past the label and into the mechanics, which is a welcome development for applicants and a mild inconvenience for anyone hoping eligibility could be resolved by vibes.
This shift reduces some of the historical daylight between the UK and the EPO, where the standard analysis for patentability of a computer-implemented invention hinges on whether the claimed subject matter produces a “further technical effect.” For AI in particular, EPO practice often turns on whether the claim is credibly tied to a technical purpose and technical effects, such as improved image processing, reduced latency, improved signal quality, or control of a technical process. That does not mean “AI” equals “technical.” It means the EPO is interested in the technical problem, the technical means, and the technical result, in roughly that order.
The practical message is that, in the UK, the question “it is software” is now less of a trapdoor and more of a cue to ask: “Fine, but what is the technical contribution?” Novelty, inventive step, and sufficiency remain fully intact, and they remain perfectly capable of doing the heavy lifting where eligibility is no longer a bar. Clearing an excluded-subject-matter objection is only the first lap. The UK may be letting more AI claims into the stadium, but it still expects them to run the race.
Across the Atlantic, the landscape remains different. In the United States, eligibility continues to be shaped by Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208 (2014). Courts ask whether a claim is directed to an abstract idea and, if so, whether it contains an inventive concept sufficient to transform the claim into patent-eligible subject matter. Simply saying an AI model runs on hardware is rarely enough. U.S. practice often demands a concrete improvement to computer functionality itself, meaning an improvement in how the system operates, not merely what it computes. If the UK is currently asking, “is this excluded,” the U.S. is still asking, “is this abstract,” which is the same question, just with a different accent.
This divergence matters because Applicants do not innovate AI for one jurisdiction. They file globally. A claim strategy that is comfortable in the UK may still face turbulence in the United States, while a U.S.-focused “technical improvement” narrative may not map cleanly onto the EPO’s technical-effect framing. The era of drafting one master claim set and translating it into local dialects is fading. Modern AI patent strategy looks more like systems architecture than document localization.
For practitioners, the lesson is both strategic and structural: Draft specifications in more detail for global flexibility. A robust AI patent application should, where supportable, describe:
Technical problem and context (not just a use case)
Model architecture and training pipeline (including data preparation and constraints)
How the model is deployed in a system (compute, memory, bandwidth, latency)
Data structures and data flow (what moves where, and why that matters technically)
Performance characteristics (metrics, tradeoffs, and operational bounds)
Concrete technical effects tied to the above
Thin disclosure leaves counsel with one story and one jurisdiction’s vocabulary. Detailed disclosure preserves options and reduces the temptation to “discover” the invention during prosecution, which tends to be less persuasive the second time it is attempted.
There is also a recurring drafting theme. Separate the mathematics from the machine, then put them back together in the specification with enough detail that the reader can see the stitching. Courts and examiners are increasingly skeptical of purely functional AI claims that focus on results rather than implementation. Anchoring an algorithm to specific computing resources, defined data flows, and measurable performance characteristics strengthens technical character in essentially every venue. Concrete detail is not a drafting luxury. It is increasingly an eligibility strategy.
It is also key for practitioners to remember that positions taken in one jurisdiction can improve strategy in another. If a U.S. examiner forces a computer-functionality improvement narrative, that framing can strengthen European prosecution. If European examination pushes limitations emphasizing technical purpose or hardware interaction, those limitations can help reduce U.S. abstract-idea risk. Global coordination is no longer a best practice. It is basic hygiene.
The broader story is familiar, given the jurisdictional divide in intellectual property law. AI is stress-testing patent doctrine worldwide, and legal systems are responding unevenly. The UK has moved toward a more pragmatic recognition that neural networks and software live in physical systems and deserve serious technical analysis. Europe continues to apply its structured technical-effect inquiry. The United States continues to litigate the outer bounds of abstraction. Harmonization remains aspirational.
In practical terms, AI innovators should assume divergence, not convergence. The UK’s posture may make it a more comfortable venue for certain computer-implemented claims, particularly where the technical contribution is clearly articulated. U.S. filings may still require sharper emphasis on improvements to computing technology itself. Drafting becomes an exercise in anticipating doctrinal weather patterns, not simply describing the invention on a calm day.
AI may operate on borderless networks, but patent law remains deeply territorial. The UK has refreshed parts of its interface. Europe has been running a structured build for years. The United States continues to issue patches for Section 101. Success will depend less on choosing one approach than on drafting so the same invention can survive in all of them.