Article title
Introduction
With the proliferation of patent applications involving artificial intelligence inventions, the united states patent and trademark office (uspto) is working to find a predictable analytical method to determine whether an artificial intelligence invention is patentable or merely abstract. To this end, uspto has used a number of tools to develop policies and guidance, establish rules of interpretation and work to standardize review processes. These institutional instruments fully reflect the powers of the federal agencies within the framework of administrative law. This paper explains the initiatives taken by uspto in the context of administrative law and reviews their practical impact on artificial intellectual inventors and patent applicants。
Background
On the basis of a long-standing, jurisprudential principle (based on 35 u. S. C. § 101), abstract concepts (such as mathematical concepts and tasks that can be carried out in the human mind) are not eligible for patent protection. Artificial intelligence and machine learning have become important sources of innovation. However, artificial intelligence and machine learning inventions are often seen as abstract concepts and thus not patentable. The uspto is conducting daily reviews of the suitability of ai patent applications. While the uspto is not in a position to decide ultimately what the law on suitability should be, it is obliged to apply the existing law consistently in daily patent applications reviews。
Foundations of administrative law: institutions can and cannot act
Like all federal, state and local bodies, uspto must comply with administrative regulations, including the administrative procedure act. Compliance with administrative law can be enforced through judicial review。
The federal agencies have various tools to change rules, regulations, procedures and policies. Such institutions could, for example:
(b) to issue a basic non-binding guidance document setting out institutional positions
(a) adjudication of cases on the basis of newly formulated or amended rules, making such decisions binding on decisions of future bodies (i. E. The formulation of rules of a adjudicative nature); and
Revision of the institutional manual to guide staff action in the institutions。
In general, the conduct of the uspto is binding only on its future actions in terms of its topicality. This is due to the fact that the validity of the patent and the appropriateness of the subject matter are finally determined by the judiciary, which does not follow the conclusions of the uspto。
In the past two years, uspto has: issued guidance documents; developed adjudicative rules; and revised its agency manual. These initiatives are designed to provide (administratively) consistency and predictability in assessing the suitability of artificially intelligent themes。
Upto issued two guidance documents on the suitability of artificial intelligence, providing examples and legal interpretations。
First, uspto's example 47-49 of the guide to topical suitability, published in july 2024, introduced specific examples of artificial intelligence, showing applicants how an artificial intelligence invention is presented as a “practical application” that meets the requirements of patentability. The july 2024 guide highlighted three conditions for practical application:
(a) limiting the concept of artificial intelligence in claims to specific areas of use
To explain in the instructions how the invention will improve technology at the technical level; and
Include non-absorptionary claims limitations that reflect (i. E. Achieved by technical means) this improvement, such as neuronet processing, real-time operations, data transmission or other steps that cannot be implemented in the human brain. See, in particular, example 47 of the july 2024 guide (claim 3). In practice, the guide is useful for applicants seeking to obtain a patent that covers artificial inventions. Although the guidance example is extremely useful in the patent review process, it is clear that the july 2024 guide did not fully meet the intended effect of uspto。
Subsequently, in august 2025, uspto sent a memorandum to the patent examiner to clarify the guidelines of july 2024 and their application to the issue of the suitability of artificial intelligence themes. In addition to the other clarifications and guidance discussed in detail in this article, the august 2025 aide-memoire reiterated the possibility of subject-specificity if an artificial intelligence invention is presented in the form of a practical application. The memorandum further directs that, when considering practical application:
The explanation of technological improvements in the notes may be implicit (rather than express) as long as the technological improvements are obvious to the general technical staff who read the patent application in this field; and
Generally speaking, the limitation of the right to record artificial intelligence training is not abstract, unless the claim clearly documents a particular mathematical algorithm, such as reverse transmission or a decline in the gradient。
In addition, the august 2025 memorandum reminded patent examiners to avoid excessive simplification of the meaning of claims in the analysis process. In order to avoid oversimplification, the memorandum states that the reviewers should consider two aspects:
(a) whether the claim merely documents the idea of a solution or outcome or covers a particular solution to an issue or a particular way of achieving the desired result; and
Whether the claim uses computers or other machines only as tools to implement existing processes, or whether the claim is intended to improve existing technologies。
Memorandum of august 2025, p. 4
The guidance document was largely followed by agency staff, but not all patent examiners chose to follow the uspto guidelines on the appropriateness of artificial intelligence themes. Recently, a reviewer responded to the argument citing the august 2025 memorandum as follows:
Given that uspto was part of the executive branch and therefore could not legislate, it was not clear how the memorandum cited would be applied。
It was clear that the examiner had wrongly stated the powers of the executive branch, which in fact could make laws or even legislation under the authority of the united states congress. However, the view of the reviewers was valid and the guidance document cited was not binding law. While the reviewers may choose to disregard the guidance of uspto, under administrative law the future decisions of the uspto must follow the outcome of a recent uspto decision with precedent effect, namely the unilateral reconsideration of the dejardan case (ex parte desjardins)。
By unilaterally reviewing the rules of a adjudicative nature in the dejardam case, uprto created a binding precedent。
In fact, the executive branch can make rules by taking decisions, which is similar to the way judges create law through common law. For example, in chenery ii, the supreme court upheld a decision of the securities and exchange commission (sec), which was based only on a new rule of law provisionally established by the sec. The reason given by the court was that the formulation of rules by adjudication provided the flexibility that was not available to the commission under article 553 of the code of administrative procedure (apa) (i. E. The “notification and commentatory rule-making” clause). However, once the executive branch has completed such a rule of adjudicatory character, it must decide similar cases in a similar way, otherwise a subsequent judgement may be considered “arbitrary” or “discretionary” under article 706 (2) a of apa. In other words, the new rules of the executive branch would be binding on future decisions of that organ。
In the unilateral review of the dejardan case, the uspto appeals review panel (arp) found that the claim involving a learning model for training machines could be an appropriate subject under step 2a, step 2a proong two, of the analytical process prescribed in the patent review procedure manual (mpep). Arp explained its own manual and the case law of the federal circuit court, applying the rule that “the claim referring to improvements in computer functionality or other technical or technical areas is proprietary”. In support of the finding in the dejardan case that “the concept of abstraction is included in the field of practical application”, arp applied the rules set out in its earlier guidance document, with the following reasoning:
(a) the dejardan's instructions describe improvements to the machine learning model itself; and
The dejardan claims reflect the improvements described in the notes。
In addition, arp criticized uspto, whose previous review of the dejardan case had analysed claims at a highly general level, equating machine learning with “non-patent algorithms” and treating the remaining additional elements as “general computer components”。
The practical effect of the dayaldan case is that patent examiners must conduct subject-specific analyses of artificial intelligence inventions in a manner similar to the dayaldan case. Unlike earlier guidance documents, the patent examiner could not say that the examples provided in the dejardan case were not binding laws, as it did (at least for the patent examiner)。
As a result of a recent amendment to mpep, it is likely that the applicant will benefit from the practical effects of the decision on patent review。
Finally, uspto has announced its intention to revise the reviewer's manual mpep and to require that the review be conducted in a manner consistent with the unilateral review of the dejardan case。
Although mpep does not have much (or any) weight before the court, it is a manual that patent examiners are required to follow. Therefore, the revision of mpep to include examples and rules of interpretation in the dejardan case is likely to have important practical implications for daily patent reviews。
On 5 december 2025, charles kim, deputy director general of patents, announced “an advance notice of a revision of mpep based on the unilateral reconsideration of the dejardan case”. This advance notice amended mpep (with immediate effect) by adding a clear directive to the patent examiner to take into account the views of the dejardan case when undertaking the article 101 thematic appropriateness analysis. In particular, mpep now makes it clear that the reviewers are instructed to look for practical applications with “improvements to other technologies or areas of technology” and cautions the reviewers “not to reject, without adequate explanation, the technical restrictions that may be relevant when assessing claims so broadly”
Conclusions
Through the use of administrative law tools, uspto has established a methodology to identify artificial intelligence inventions as subject-specific, based on existing case law on patent suitability. It is reasonable to expect that the patent examiner will follow this approach in the near future in the review of patent applications. The current system is beneficial to artificial innovators and patent applicants. (compiled from www. Mondaq. Com)
Narrator:
