A more open approach to AI patenting in the US

Recent developments at the USPTO will make it easier to secure US patent protection for AI technologies. Director Squires’ first few months at the USPTO are expected to have a very significant impact on the assessment of eligibility for computer-implemented inventions including AI. As European patent attorneys, we are watching these changes with interest and satisfaction that the USPTO’s rapid-fire revisions appear to be moving its practice closer to European practice, such that new and inventive solutions to technical problems will usually not be excluded from patent protection merely because they are implemented in software.

Recent developments at the USPTO

USPTO Director Squires has changed examination and appeals practice for AI inventions under 35 USC §101, by a sequence of actions that remind his Examiners of existing case law and discourage over-reliance on §101 when examining patent applications.

In the 26 September 2025 decision of the PTAB Appeals Review Panel (ARP) in Ex parte Appeal 2024-000567^[1] (Desjardins), Director Squires and colleagues provided a very clear message that computer-implemented methods of training a machine learning model are eligible for patent protection under §101, and that the assessment of patentability for such methods should be based on §§102, 103 and 112 of the US code. The ARP agreed with the Appellant’s assertion that the claims integrated a mathematical processing step (“computing… an approximation of a posterior distribution over possible values of the plurality of parameters”) into a practical application, accepting the Appellant’s explanation that the claimed invention addresses the technical problem of ‘catastrophic forgetting’ in continual learning systems to preserve performance on a first machine learning task when training of a machine learning model for a different machine learning task.

In brief, the Desjardins ARP decision reminds us that we must consider the claims as a whole and consider whether any abstract ideas (such as mathematical claim features) have been integrated into a practical application, while noting that:

“claims directed to an improvement in the functioning of a computer, or an improvement to other technology or technical field are patent eligible”

(See MPEP §§ 2106.04(d)(1) and 2106.05(a), both of which were revised on 5 December and which cite Enfish, LLC v. Microsoft Corp., 822 F.3d 1327, 1339 (Fed. Cir. 2016)).

Referring to Enfish as one of the Federal Circuit’s leading cases on eligibility of technological improvements, the ARP decision notes that Enfish recognized that:

“[m]uch of the advancement made in computer technology consists of improvements to software that, by their very nature, may not be defined by particular physical features but rather by logical structures and processes” (822 F.3d at 1339)

and that

“[s]oftware can make non-abstract improvements to computer technology, just as hardware improvements can“.

The ARP decision notes that the eligibility determination should turn on whether:

“the claims are directed to an improvement to computer functionality versus being directed to an abstract idea.” (Id. at 1336).

Barely two months after the Desjardins decision of the ARP, with the decision made precedential by the USPTO on 4 November, US patent attorneys are already reporting allowance of some AI-related patent applications that previously faced objections. Despite the USPTO insisting that its August memo on application of §101 was merely a reminder of correct practice and that the ARP’s Desjardins decision reflects existing case law, this appears to be a significant shift in USPTO examination practice. The Desjardins ARP decision encourages a narrower interpretation of which patent claims should be considered to be “directed to” an abstract idea or other patent-ineligible subject matter, making it easier for Examiners to approve claims on the basis that a mathematical method that achieves a technological improvement is integrated into a practical application.

Removing any doubt about whether USPTO examiners need to pay attention to the ARP’s decision, the USPTO Deputy Commissioner for Patents issued a memorandum to the Patent Examining Corps on 5 December quoting from the ARP decision and amending the Manual of Patent Examining Practice (MPEP)[2]. The Director has even issued a memo offering applicants and patent practitioners a tip for getting cases past his examiners, via a Subject Matter Eligibility Declaration (SMED)[3]. It remains to be seen how much use is made of this option, but it is worth considering for tricky cases.

The clarity and simplicity of the ARP’s comments and the rapid inclusion in the MPEP will help USPTO examiners to apply the ARP’s guidance as they work their way through the USPTO’s complex multi-prong test.

For observers in Europe, it is easy to agree with Director Squires that solutions to technical problems should be eligible for patent protection. The ARP’s adjustment of USPTO practice with its refocussing on 35 USC §§ 102, 103 and 112 and reduced focus on §101 should bring USPTO practice closer to EPO practice, at least on the fundamental question of whether a claimed invention is excluded from patent protection. Also, the comments in Enfish that were approved by the ARP align closely with historic comments from some of the UK’s leading patent judges. Both of these are discussed below. Progress towards international harmonisation of patent laws is usually good for industry leaders, regardless of whether their head office is in California, Colorado or the UK (where DeepMind was founded), because inconsistent laws in their major markets increase costs and can result in a very different scope of protection being obtained in different countries.

Of course, neither August’s reminder of the correct practice for USPTO examination art units or the new precedential decision of the Appeals Review Panel of the USPTO’s PTAB constrain US courts or legislators, who may express a different view. Some US commentators suggest that US courts are struggling to consistently apply §101 to AI/ML inventions. Perhaps with this in mind, Director Squires gave a statement to the US Senate Judiciary Subcommittee on Intellectual Property on 9 October to encourage an expansive approach to the assessment of eligibility. His statement^[4] notes that:

“The drafters of our patent laws [….] chose deliberately broad terms—process, machine, manufacture, composition of matter—so that the law could accommodate technological revolutions unforeseen in their time. Just as their words encompassed the telegraph, the telephone, and the airplane, so too must they encompass the blockchain, the quantum processor, and the diagnostic algorithm”

The Director summarised the Ex parte Desjardins case and reminded the Subcommittee of his own recent statement that: “crypto and AI to quantum computing and diagnostics […] are applied, patent-eligible technologies driving the frontiers of knowledge”  and stated that “I want inventors and entrepreneurs to know: the USPTO is open for business—not only for the technologies of today, but especially for those of tomorrow.”

Advocating an expansive approach to patent eligibility for economic and strategic policy reasons, Director Squires highlighted the importance of some specific technical fields:

“Artificial intelligence, quantum computing, clean energy, advanced materials—these are the arenas in which the next century’s balance of power will be determined.”

The Director’s call for reform of 35 USC §101 seems to align closely with previous statements from the Chair of the Senate Judiciary Subcommittee, Senator Thom Tillis, who has been working alongside other Senators for some years to steer a course towards reform of §101. It will be interesting to watch progress of the Patent Eligibility Restoration Act 2025 now that these two strong voices are so aligned.

The clarity of Director Squires’ message, and recognition within the Senate Judiciary Subcommittee of the importance of certainty and predictability for innovators and investors, increase the likelihood of positive change. As US legislators consider their options for replacing 35 USC § 101, spurred on by this strong encouragement from Director Squires and many other respected voices in the US IP community, they might wish to consider these points:

(1) the benefits to US-based multinationals of international harmonisation of IP laws;

(2) how predictable patent eligibility is at the European Patent Office thanks to Europe’s test being easier to apply than the current complex eligibility test that USPTO examiners have to apply when assessing §101. The EPO approach is summarized below.

(3) that a test for patentability that focusses on whether a claimed invention solves or at least mitigates a technical problem and focusses on the inventors’ contribution to the art would be better than an eligibility test which guides examiners to scan the claims looking for a potentially problematic isolated feature and only subsequently assesses whether the identified “problem” feature is integrated into a practical application. Wouldn’t it be better to start with an assessment of the invention as a whole, with consideration of the technical problems, how the invention works to address those problems and its advantages? With that more positive starting point, after recognizing that the claim as a whole is directed to a technological solution to a technical problem, examiners and Courts can quickly move on to §§102, 103 and 112. There is often no need to assess whether an isolated feature is “abstract” or mathematical if the claim as a whole is recognised as being directed to a practical application or otherwise addresses a technical problem.  The ARP’s Desjardins decision on §101 got to the right result on eligibility, but a simpler eligibility test may avoid the PTAB ARP having to “step in” in future.

Patent attorneys should take note of Director Squires’ comment that “§§ 102, 103 and 112 are the traditional and appropriate tools to limit patent protection to its proper scope”.

These requirements for novelty, inventive step and sufficiency of description remain very significant “hurdles” for innovators and their patent attorneys and function well as limitations on patent scope, including in rapidly developing complex technologies such as AI and quantum computing.

So we support Director Squires’ encouragement to confer patentability on the “advancement made in computer technology [which] consists of improvements to software… defined… by logical structures and processes”, if the claimed invention is new, inventive and described in sufficient detail to enable a skilled person to put the invention into effect.

Only two months after Desjardins, the USPTO has reconsidered another significant issue for AI patenting – withdrawing its February 2024 “Inventorship Guidance for AI-Assisted Inventions” and issuing revised guidance^[5]. Reminding us of the Federal Circuit’s conclusion that “AI cannot be named as an inventor”, and that “conception” is “the touchstone of inventorship”, the USPTO’s revised inventorship guidance tells us that “The same legal standard for determining inventorship applies to all inventions, regardless of whether AI systems were used in the inventive process. There is no separate or modified standard for AI-assisted inventions”, and that “The Pannu factors only apply when determining whether multiple natural persons qualify as joint inventors”. Fundamentally, the new USPTO guidance points us back to the case law of the Federal Circuit and points out that

“Determining inventorship is highly fact intensive” and that the USPTO will now presume that the inventors named on the application data sheet or oath/declaration are the actual inventors of the application. In other words, these case-specific fact-intensive determinations will be left for the courts.

This is the same pragmatic approach as taken in other countries: only human inventors can be named, and must be named, but it is not the role of patent offices to check the specific contributions of human inventors when using AI systems or whether the human inventors used an AI system.

The determination of inventorship will not be easy for the courts, and every company needs to develop an AI usage policy which makes clear which specific AI-based tools are approved for which purposes, and to develop a data retention policy including whether to retain records of the specific contributions of each human inventor, so they are ready for potential future validity challenges based on an alleged lack of human inventorship of AI-assisted inventions. As well as developing AI usage policies, any company that values its IP must educate its R&D teams about the potential risk of unintended disclosure of an invention as soon as it is devised, if using a generative AI tool to explore and extend a new invention without enterprise data protection that blocks training of the underlying LLM.