Patentability of AI inventions in the UK US and China

As AI continues to advance at pace, securing patent protection for AI‑related inventions has become increasingly important, particularly for maintaining competitive advantage and attracting investment.

But as with many things relating to AI and the law, a common international approach has not emerged.

Many jurisdictions broadly agree that an AI system itself cannot be named as an inventor, but no unified framework exists on the patentability of inventions embodying AI technologies, such as a recommendation system based on an Artificial Neural Network (ANN).

This is our Q2 2026 snapshot to show where the law is heading in the three key territories of the UK, US and China.

Patentability of AI inventions in the UK

On 11 February 2026, the UK Supreme Court handed down judgment in Emotional Perception AI Ltd v Comptroller-General of Patents [2026] UKSC 3. This gives much-needed clarity on how patents relating to AI will be examined in the UK and may mark the dawn of an increasingly AI-friendly era.

The case involved an appeal against the UKIPO's decision to refuse a patent based on the statutory exclusion from patentability for "a program for a computer…as such."

The patent application claimed an improved system for providing media file recommendations to an end user, such as music tracks which a particular user might enjoy based on existing preferences. The system is powered by two trained ANNs.

Previously in the UK, to be patentable, a computer-based invention had to have a "technical effect" so that it amounted to more than a computer program "as such" and therefore wasn't caught by the exclusion from patentability.

This meant the invention had to be directed to a specific technical process and contribute to a solution of a technical problem lying outside of the computer.

For the past 20 years, this has been assessed by the UK courts by reference to the Aerotel test, which involved consideration of the problem to be solved by the alleged invention, how the invention works, what the advantages are and what the inventor has added to human knowledge. 

This is no longer the case. In assessing whether there are technical features that take the invention outside of the exclusion for "computer programs as such," the Supreme Court decided that the Aerotel test should no longer be followed; instead, it adopted an approach more in line with that taken by the European Patent Office.

This "any hardware" approach assumes that the relevant technical means are present because computer hardware is needed to implement the ANN, so the ANN is not a program for a computer "as such."

The result of the judgment is that AI inventions will now be considered in line with the usual requirements for patentability of any invention (they must be new, inventive and capable of industrial application), subject to a new "intermediate step" before that analysis.

In this intermediate step, features of the invention that do not contribute to, or interact with, the technical character of the invention viewed as a whole are filtered out from the assessment of patentability, so they do not play a part in consideration of inventive step.

The function of this intermediate step is, therefore, concerned with excluding features of the invention from subsequent consideration, rather than with determining the patentability of the invention as a whole.

The judgment may lead to a reworking of the UKIPO’s guidelines for examining patent applications relating to AI – watch this space.

For more insight on the Supreme Court's decision, read our analysis.

Patentability of AI inventions in the US

Recent guidance from the USPTO has reinforced how AI-related inventions should be examined under Reminders on evaluating subject matter eligibility of claims under 35 U.S.C. 101, which is articulated in the in the Manual of Patent Examining Procedure (MPEP).

The current framework includes a determination of whether a claim recites an abstract idea, which is limited to three groupings:

1. mathematical concepts;
2. certain methods of organizing human activity; and
3. mental processes.

One of the examples given by the USPTO highlights the distinction between a claim reciting a judicial exception versus a claim merely involving a judicial exception.

Examiners typically rely on the mental process grouping when issuing subject matter eligibility rejections for AI-related claims.

The USPTO emphasized that claims for AI-related inventions that “encompass AI in a way that cannot be practically performed in the human mind do not fall within this grouping.”

If the examiner determines that the claim cannot be practically performed in the human mind and no other judicial exception applies, the claim is eligible.

But if the examiner determines that any of the judicial exceptions apply, further eligibility analysis is required.

For further analysis, the examiner must “evaluate whether the claim as a whole integrates the recited judicial exception into a practical application.”

While there are other considerations, the USPTO focused on two key considerations for AI-related inventions: improvements to technology and mere instructions to apply an abstract idea (“apply it”).

A recent decision from the Appeals Review Panel of the USPTO, Advance Notice of Change to the MPEP in Light of Ex Parte Desjardins, USPTO, 1 (Dec. 5, 2025),  elaborated on the consideration of improvements to technology.

The decision cited Enfish, LLC v. Microsoft Corp., 822 F.3d 1327, 1336 (Fed. Cir. 2016)., which states that “software can make non-abstract improvements to computer technology, just as hardware improvements can.” The software improvements may be improvements in “logical structures and processes.”

The Advance Notice of Change to the MPEP in Light of Ex Parte Desjardins states that “when evaluating a claim as a whole, examiners should not dismiss additional elements as mere ‘generic computer components’ without considering whether such elements confer a technological improvement to a technical problem, especially as to improvements to computer components or the computer system.”

In the Reminders on evaluating subject matter eligibility of claims under 35 U.S.C. 101, the USPTO noted that “the ‘apply it’ consideration often overlaps with the improvements consideration.” If the AI-related claim is integrated into a practical application, the claim is patent eligible.

Patentability of AI inventions in China

Two relevant documents here are the Patent Examination Guidelines and the Guidelines for Patent Applications Related to Artificial Intelligence (Trial) issued by the China National Intellectual Property Administration (CNIPA).

These confirm that the subject matter of AI-related inventions must provide a technical solution that employs technical means to solve a technical problem and achieve a technical effect in accordance with the laws of nature, which excludes abstract mathematical theories, algorithms and computer programs.

The CNIPA gives three example scenarios where an AI-related invention may qualify as a patentable technical solution:

• the AI model/algorithm processes data with specific technical meaning in a technical field and achieves a technical effect;
• the AI model/algorithm is technically tied to the internal architecture of a computer system and improves computational efficiency or execution performance; and
• the AI model/algorithm discovers intrinsic correlations consistent with natural laws in data and improves the reliability or accuracy of big data analysis for a specific application.

The 5 December 2023 CNIPA Patent Re-examination Decision No. 1489791 regarding Chinese patent application No. 201611262293.6 gave a practical example of the first scenario.

The application concerned a system and method for compressing a Recurrent Neural Network that received specific input data, such as images, internet resources, documents, personalized recommendations for users, and text or spoken utterances.

The CNIPA held that the target to be processed in the patent application was data with specific technical meaning in the technical field, and that the solution achieved technical effects such as improved data processing effectiveness while using less data storage. As such, the claimed solution constituted a patentable technical solution.

More recently, a 13 November 2025 amendment to the Patent Examination Guidelines also emphasizes Article 5(1) of the Chinese Patent Law in the context of AI-related inventions. AI-related inventions will be rejected if data collection, label management, rule configuration, or recommendation decision-making contravenes laws, social morality or public interests.

This rejection rule would, for example, apply to an autonomous vehicle decision model that selects who to protect based on gender and age.

Inventorship: The shared prohibition on AI as inventor

Despite the differences in approach outlined above, the UK, the US and China each agree on one point: that AI systems cannot be named as inventors on patent applications, and that only natural persons may qualify as inventors. 

This has been tested most prominently through Dr Stephen Thaler's attempts to register his AI system, DABUS, as the sole inventor on various patent applications.

Importantly, none of these jurisdictions have ruled out patent protection for inventions in which AI played a significant role as a tool in the inventive process, if a human being can be identified as having made the relevant inventive contribution.

Practical considerations on patenting AI inventions in the UK, US and China

The law relating to AI and patents is likely to evolve worldwide. For now, applicants seeking protection across multiple jurisdictions, including the UK, US and China, should take note of these practical points:

• Businesses should ensure clear documentation of human contributions throughout the development process to satisfy the inventorship requirements applicable in all three jurisdictions. This is particularly important, given the common position that AI systems cannot be named as inventors, and that a natural person must be identified as having made the relevant inventive contribution.
• Claims should be drafted to emphasize the technical problems solved and the technical effects achieved, rather than focusing solely on the AI algorithm itself. This approach is relevant in all three jurisdictions: in the UK, the new intermediate step filters out non-technical features; in China, the "three-part" technical solution test demands a concrete technical effect; and in the US, integration into a practical application or demonstration of a technological improvement is required to overcome the abstract idea bar.
• Given the variations between jurisdictions, businesses with international patent portfolios should develop filing strategies tailored to each jurisdiction's specific requirements while maintaining consistency where approaches have converged, particularly between the UK and EPO following the Emotional Perception judgment.
• Applicants filing in China should be alert to the additional ethical dimension introduced by the 2025 amendment to the Patent Examination Guidelines, which may result in rejection where data collection or decision-making processes contravene public interests or social morality. This represents a distinctive feature of the Chinese regime that has no direct equivalent in the UK or US frameworks.

If you have questions on AI, patents or anything else, please get in touch with the authors or your regular DLA Piper contacts.