Written by: Charles Dresser

Introduction 

With many patent applications covering AI inventions, the patent office has sought to establish predictable analysis methods for determining whether an AI invention is patent-eligible or an abstract idea. To this end, the patent office has employed a number of agency tools to create policy and guidance, establish interpretive rules, and formalize the examination process. These agency tools are informative of the powers of federal agencies under administrative law. This paper explains the efforts taken by the patent office within the context of administrative law, while opining on the practical consequences of these efforts for the AI inventor and patent applicant. 

This paper is the third in a series on this subject, including: 

1. AI Patent Strategies Following the USPTO’s 2024 Guidance 
2. USPTO AI Patent Eligibility: The 2025 Examiner Memo 

Background

Abstract ideas (e.g., mathematical concepts and tasks that may be performed in a human mind) are not eligible for patent protection, through a longstanding judicially created doctrine under 35 U.S.C. § 101. Artificial intelligence (AI) and machine learning (ML) have become sources of innovation. But, AI and ML inventions can be considered abstract and found ineligible for patent protection. The USPTO examines AI patent applications every day for eligibility. While the USPTO is not responsible for deciding what the ultimate law is on eligibility, the USPTO is responsible for consistently applying current eligibility law to day-to-day patent application examination. 

Administrative Law Primer: What Agencies May—and May Not—Do 

The USPTO, like all federal, state, and local agencies, must comply with administrative law, including the Administrative Procedure Act (APA). Compliance with administrative law may be enforced through judicial review. 

Federal agencies have a number of tools to change rules, regulations, procedures, and policies.ⁱ For example, agencies may: 

(1) publish largely non-binding guidance documents, indicating agency positions;  

(2) adjudicate cases with new or amended rules, such that the cases bind future agency decisions (i.e., adjudicatory rulemaking), see SEC v. Chenery Corp., 332 U.S. 194 (1947) (“Chenery II”); and  

(3) amend agency manuals that instruct the actions of agency employees. 

Generally, with regard to subject matter eligibility, the USPTO’s actions will only bind future USPTO actions. This is because patent validity and subject matter eligibility are decided by the judiciary with no deference to USPTO findings. 

Over the past two years, the USPTO has (1) published guidance documents; (2) engaged in adjudicatory rulemaking; and (3) implemented changes in its agency manual aimed at providing (administrative) consistency and predictability in AI subject matter eligibility evaluation. 

(1) The USPTO published two guidance documents providing examples and legal interpretation on AI eligibility. 

First, the USPTO’s July 2024 Subject Matter Eligibility Guidance Examples 47–49 introduced AI-specific examples showing applicants how an AI invention can be presented to be patent-eligible “practical application.”ⁱⁱ The July 2024 Guidance emphasized three conditions for recognition of a practical application: (1) limit the AI concept to a particular field of use within the claims, (2) describe in the specification a technical explanation of how the invention improves technology, and (3) include non-abstract claim limitations that reflect (i.e., technologically implement) that improvement—such as neural-network processing, real-time actions, data transmission, or other steps that cannot be performed mentally. See July 2024 Guidance, Example 47, claim 3. Practically, the July 2024 Guidance has been advantageous for applicants seeking a granted patent covering AI inventions. Details of the practical consequences of the July 2024 Guidance may be found in this article dated May 2025. While guiding examples can be tremendously useful in patent prosecution, the July 2024 Guidance did not, apparently, achieve the complete effect desired by the USPTO.  

In August 2025, the USPTO published a memorandum to patent examiners to clarify the July 2024 Guidance and its application to AI subject matter eligibility questions. Among other clarifications and guidance discussed in detail in this article, the August 2025 Memo reinforced that an AI invention may be subject matter eligible if presented as a practical application. The August 2025 Memo further instructed that, when considering a practical application: 

1. the specification’s explanation of technological improvement may be implicit (rather than explicit), as long as the technological improvement is manifest to a person of ordinary skill in the art reading the patent application; and 
2. claim limitations reciting AI training are, generally, non-abstract, unless the claim expressly recites specific mathematical algorithms, such as backpropagation or gradient descent. 

Additionally, the August 2025 Memo cautions patent examiners from oversimplifying the meaning of the claims during analysis. To avoid oversimplification, the August 2025 Memo reiterated that the examiner should consider two considerations:  

1. Whether the claim recites only the idea of a solution or outcome. . . or the claim covers a particular solution to a problem or a particular way to achieve a desired outcome [; and]   
2. Whether the claim invokes computers or other machinery merely as a tool to perform an existing process, or whether the claim purports to . . . improve an existing technology (emphasis added). 

August 2025 Memo, p. 4. 

Guidance documents are largely followed by agency employees, but not all patent Examiners have chosen to follow the USPTO’s AI subject matter eligibility guidance. We recently received the following from an Examiner in response to arguments citing the August 2025 Memo: 

“it is unclear how to apply the cited Memo given . . . the [Patent] Office is part of the executive branch and, as such, cannot make law.”ⁱⁱⁱ 

The Examiner misstates the powers of the Executive branch, which may make law and even legislate when delegated that power by Congress. But the Examiner’s point is valid—the cited guidance document is not binding law (and you can’t make me follow it). While the Examiner may be free to ignore USPTO guidance, under administrative law, future USPTO decisions must accord with a recent precedential USPTO adjudication, ex parte Desjardins. 

(2) The USPTO created binding precedent through adjudicatory rulemaking in ex parte Desjardins. 

An agency may engage in rulemaking through adjudication, akin to the way judges make law through the common law method. For instance, in Chenery II, the Supreme Court upheld a decision by the Securities and Exchange Commission (SEC) that rested only on the SEC’s ad hoc formulation of a new legal rule.^iv The court reasoned that rulemaking through adjudication afforded the SEC the needed flexibility not available through notice and comment rulemaking, under APA § 553. However, once an agency makes a decision, through adjudicatory rulemaking, the agency must adjudicate similar cases to similar holdings, else the later judgments be held arbitrary or capricious under APA §706(2)(A). Said another way, the new agency rule is binding on future agency decisions.^v 

In ex parte Desjardins (Desjardins), the USPTO Appeals Review Panel (ARP) held claims directed to training a machine learning model subject matter eligible under Step 2A Prong Two of the MPEP prescribed analysis process. Appeal No. 2024-000567, Application 16/319,040, ARP Decision on Request for Rehearing, p. 7 (P.T.A.B. Sept. 26, 2025) (Precedential, Nov. 4, 2025). The ARP interpreted its own manual and Federal Circuit jurisprudence, applying the rule that “claims directed to an improvement in the functioning of a computer or an improvement to other technology or technical field are patent eligible.” Id., p. 8 (citing MPEP §§ 2106.04(d)(1); 2106.05(a) (citing Enfish, LLC v. Microsoft Corp., 822 F.3d 1327, 1339 (Fed. Cir 2016) and McRo, Inc. v. Bandai Namco Games Am. Inc., 837 F.3d 1299, 1315 (Fed. Cir. 2016))). To support the ruling that Desjardins incorporated an abstract idea into a practical application, the ARP, applying the rules espoused by its earlier guidance documents, reasoned that:  

1. Desjardins’s specification described an improvement in training the machine learning model itself;^vi and  
2. Desjardins’s claims reflected the improvement from the specification.^vii 

Additionally, the ARP faulted previous USPTO examination of Desjardins for analyzing the claims at a high level of generality, in which machine learning was equated with “unpatentable “algorithm, and the remaining additional elements as “generic computer components.” Desjardins, p. 9.  

The practical impact of Desjardins is that patent examiners must perform subject matter eligibility analysis of AI inventions similarly to Desjardins. Unlike earlier guidance documents, patent examiners cannot say that the example provided by Desjardins is not binding law, because (at least for patent examiners) it is. 

This ruling’s practical effects on patent examination are likely to be enjoyed by applicants because of a recent amendment to the Manual of Patent Examination and Process (MPEP). 

(3) Finally, the USPTO has announced intended amendments to the Examiner’s manual, MPEP, to require examination consistent with ex parte Desjardins.  

While the MPEP does not carry much (if any) weight with courts, it is the manual that patent examiners are expected to follow. Accordingly, amending the MPEP to include the example and interpretive rules of Desjardins is likely to have great practical importance in day-to-day patent examination. 

On December 5, 2025, Charles Kim, Deputy Commissioner for Patents, announced an “Advanced notice of change to the MPEP in light of Ex parte Desjardins.” C. Kim, Advanced notice of change to the MPEP in light of Ex parte Desjardins, USPTO (December 2025). This advance notice revised the MPEP (effective immediately) to include explicit instructions to patent examiners to consider the Desjardins opinion when performing 101 subject matter eligibility analysis. Specifically, the MPEP now plainly (1) instructs examiners to search for a practical application with “an improvement to other technology or technical field”; and (2) cautions examiners “not to evaluate claims at such a high level of generality that potentially meaningful technical limitations are dismissed without adequate explanation.” MPEP § 2106.04(d)(1).  

Conclusion 

Using administrative law tools, the USPTO has established a consistent approach for finding AI inventions subject matter eligible, under current patent eligibility jurisprudence. Patent examiners can be reasonably expected to follow this approach in patent application examination for the near future. This current regime is advantageous for AI inventors and patent applicants. 

This publication is distributed with the understanding that the author, publisher, and distributor of this publication and/or any linked publication are not rendering legal, accounting, or other professional advice or opinions on specific facts or matters and, accordingly, assume no liability whatsoever in connection with its use. Pursuant to applicable rules of professional conduct, portions of this publication may constitute Attorney Advertising. The choice of a lawyer is an important decision and should not be based solely upon advertisements.

---

Endnotes

ⁱ Some of these changes to the USPTO’s stance on AI are thought to be attributable to a change in presidential administration. So long as the changes are within the bounds of administrative law, the Supreme Court views agency policies that change with the President as a part of the democratic process. See Motor Vehicles Manufacturers Ass’n v. State Farm Mutual Automobile Insurance Co., 463 U.S. 29, 57 (1983) (J. Rehnquist concurring in-part and dissenting in-part “[t]he agency’s changed view of the standard seems to be related to the election of a new President of a different political party. It is readily apparent that the responsible members of one administration may consider public resistance and uncertainties to be more important than do their counterparts in a previous administration. A change in administration brought about by the people casting their votes is a perfectly reasonable basis for an executive agency’s reappraisal of the costs and benefits of its programs and regulations. As long as the agency remains within the bounds established by Congress, it is entitled to assess administrative records and evaluate priorities in light of the philosophy of the administration”). 

ⁱⁱ Eligibility is not (technically) merely matter of how the AI invention is presented—the AI invention needs to actually represent an improvement in (AI) technology. This point is clarified by recent Federal Circuit opinion, Recentive Analytics, Inc. v. Fox Corp., in which the court held that: “patents that do no more than claim the application of generic machine learning to new data environments, without disclosing improvements to the machine learning models to be applied, are patent ineligible under § 101.” Case No. 2023-2437 (Fed. Cir. Apr. 18, 2025). Nevertheless, an AI invention that does represent an improvement in AI technology will be found ineligible if it fails to be presented as such. 

ⁱⁱⁱ The Examiner continued, “the Memo by its own terms does not apply to Technology Center 3700, which is where the instant Application is being examined, but instead only applies to [Technology Center]’s 2100, 2600, and 3600.” The August 2025 Memo, which was published to the world, does recite “TO: Technology Centers 2100, 2600, and 3600” at the top of the first page. One may characterize this part of the Examiner’s argument as “I never got the memo.” 

^iv The new legal rule was within SEC’s statutory authority to create, under §§ 7 and 11 of the Holding Company Act. 

^v An agency may, of course, change its rule or policy, but only after articulating a rationale for doing so. For example, in FCC v. Fox, the Supreme Court held that FCC’s change in policy was not arbitrary or capricious, under APA § 706, because the FCC had indicated the departure from earlier practices and provided reasonable bases for the new policy. 556 U.S. 502 (2009). The FCC had official policy that precluded liability for broadcasters, under 18 U.S.C. § 1464, for fleeting expletive use of the words “FUCK” and “SHIT,” when the word is used only once. Notwithstanding established office policy, on March 15, 2006, the FCC sent Notices of Apparent Liability for two separate expletive usages during a Fox broadcast where the words were used only once. At the 2003 Billboard Music Awards, Cher said “I’ve also had critics for the last 40 years saying that I was on my way out every year. Right. So fuck‘em” and Nicole Richie said, in a separate segment, “Why do they even call it ‘The Simple Life?’ Have you ever tried to get cow shit out of a Prada purse? It’s not so fucking simple.” The Notices of Apparent Liability recognized FCC’s previous regime of not finding liability for fleeting expletive use, provided reasons for the change (the categorical exemption of liability would likely lead to more widespread usage and technological advances have made it easier for broadcasters to beep out expletives), and provided a new policy statement to replace that which was being abandoned (single use of the expletive only “weigh[s] against a finding of indecency”). 

^vi At [0021] of the Specification, “allow[ing] artificial intelligence (AI) systems to “use less of their storage capacity” and enable[ing] “reduced system complexity,” Desjardin, p. 9. 

^vii Claim 1 . . . reflects the improvement: “adjust the first values of the plurality of parameters to optimize performance of the machine learning model on the second machine learning task while protecting performance of the machine learning model on the first machine learning task, id.