November 'USPTO Hour' on Subject Matter Eligibility Precedent
Webinar on §101 Updates
The United States Patent and Trademark Office (USPTO) recently convened the “USPTO Hour” to address the state of subject matter eligibility under 35 U.S.C. § 101. Hosted on November 5, 2025, the session featured high-level officials including Deputy Commissioner for Patents Charles Kim, Vice Chief Administrative Patent Judge Michael Kim, and Senior Legal Advisor Carolyn Kosowski. The video was posted to YouTube recently, but does not include guidance on the SMEDs memos issued on December 4, 2025.
The presentation focused on three primary developments: the August 2025 “Reminder” Memorandum, the precedential Appeals Review Panel (ARP) decision in Ex parte Desjardins, and the Delegated Rehearing Panel (DRP) decision in Crusoe Energy System, LLC v. UpStream Data Inc. (PGR2023-00039).
For patent practitioners and stakeholders, the message was distinct: the USPTO is signaling a shift toward eligibility for technical improvements in software and Artificial Intelligence (AI), while reinforcing that §§ 102, 103, and 112 remain the primary tools for limiting patent scope.
The August 2025 Memorandum: A “Reminder” on Procedure
Deputy Commissioner Charles Kim noted that while § 101 rejection rates dropped significantly following the 2019 guidance, rates have recently “increased in certain TCs” (Technology Centers), specifically TCs 2100, 2600, and 3600 (Presentation, pp. 5, 6).
In response, the USPTO issued a memorandum to the entire examining corps on August 4, 2025 (p. 6).
Distinguishing Mental Processes from AI Operations
A critical takeaway for software practitioners involves the “mental process” grouping of abstract ideas. The presentation highlighted that terminology used to describe AI functions often mimics human cognition, but this should not lead to a § 101 rejection (p. 8)6.
“Claim limitations that encompass AI in a way that cannot be practically performed in the human mind do not fall in the mental process grouping” (p. 8).
Examiners were explicitly instructed not to expand this grouping to cover AI limitations that are practically impossible for a human to perform, such as complex neural network processing (p. 8).
Kosowski elaborated during the webinar:
“For example, we say that large language models can hallucinate when they produce nonsensical or inaccurate information, but that is not the same as a human being experiencing hallucinations... Examiner[s] should only find that a claim limitation recites a mental process when that limitation can reasonably be performed in the human mind” (video, 13:27).
“Reciting” vs. “Merely Involving”
The panel clarified the distinction between claims that “recite” a judicial exception and those that merely “involve” one (p. 9).
A claim recites an exception if it explicitly sets forth or describes a law of nature or abstract idea (p. 9).
Recites: “A machine comprising elements that operate in accordance with F=ma” (p. 10).
Merely Involves: “A teeter-totter comprising an elongated member pivotably attached to a base member...” (which relies on the law of the lever but does not recite the formula) (p. 10).
Preponderance of Evidence
Perhaps most importantly for prosecution strategy, the memo reminds examiners that uncertainty is not a valid basis for rejection. “Claims should not be rejected simply because an examiner is uncertain about the claim’s eligibility” (p. 15).
If it is a “close call,” the examiner’s rejection should not be made, as unpatentability must be established by a preponderance of the evidence (p. 15).
Ex parte Desjardins: A Victory for AI Training Models
The webinar highlighted the Ex parte Desjardins decision, which was designated precedential. This decision is significant for AI innovators as it validates that claims to AI inventions reflecting an improvement to technology are eligible (p. 30).
The Invention
The application (No. 16/319,040) addressed “catastrophic forgetting” in machine learning—a phenomenon where models lose knowledge of a previous task when training on a new one (slides, p. 18).
The claims recited a method of training that included an “adjust” limitation:
“...adjusting the first values of the plurality of parameters to optimize an objective function that depends in part on a penalty term that is based on the determined measures of importance of the plurality of parameters to the first machine learning task” (p. 19).
The Decision
The Appeals Review Panel (ARP) vacated the § 101 rejection (p. 21). The presentation noted that the panel criticized the lower board’s reasoning:
“[U]nder the panel’s reasoning, many AI innovations are potentially unpatentable—even if they are adequately described and nonobvious—because the panel essentially equated any machine learning with an unpatentable ‘algorithm’ and the remaining additional elements as ‘generic computer components,’ without adequate explanation” (p. 23).
The ARP found that while the claims involved mathematical concepts, they reflected a specific “improvement to the functioning of a computer” (p. 22).
The decision emphasized that “software can make non-abstract improvements to computer technology, just as hardware improvements can” (p. 22; video, 31:09).
However, practitioners must note that the claims remained rejected under § 103 (obviousness). The panel emphasized that §§ 102, 103, and 112 are the “traditional and appropriate tools to limit patent protection,” and those statutes “should be the focus of examination” (p. 24).
Crusoe Energy: Burden of Proof in Post-Grant Reviews
The panel also discussed Crusoe Energy System, LLC v. UpStream Data Inc., a Post Grant Review (PGR) case (p. 25, PGR2023-00039). The initial decision found claims, for U.S. Pat. No. 11,574,372, directed to using natural gas to power blockchain mining ineligible as a “fundamental economic practice” (p. 27).
Upon review, the Delegated Rehearing Panel (DRP) vacated this finding (p. 28). The DRP reasoned that the petitioner failed to meet their burden because the independent claims did not describe a result that was itself an abstract idea, but instead “describe[] a machine defined by its constituent parts” (p. 28; video, 36:40).
This reinforces that the burden lies heavily on the challenger (or examiner) to identify specific claim language that recites the abstract idea (p. 28).
Analysis: Benefits, Challenges, and Risks
Benefits
AI Eligibility Safe Harbor: Desjardins provides a solid pathway for eligibility: claiming improvements to the training methodology or the internal functioning of the model is likely eligible under Enfish (p. 21).
Reduced Friction: The instruction to examiners to avoid “close call” rejections and the clarification on mental processes should reduce the volume of § 101 rejections for technical software inventions (p. 15).
Challenges
Examiner Consistency: Despite the memo, applying the “improvement” standard remains subjective, particularly regarding whether the specification provides sufficient details for one of ordinary skill to recognize the improvement (p. 13). Deputy Commissioner Kim acknowledged that “consistency is always a challenge” when dealing with over 8,000 examiners (video, 47:11).
The “Apply It” Trap: The memo warns examiners against oversimplifying the “apply it” consideration. Applicants must ensure their specifications detail how the improvement is achieved to avoid falling into the “result-oriented” trap (p. 14).
Risks
Obviousness Remains a Hurdle: As seen in Desjardins, overcoming § 101 does not secure the patent. The Office is explicitly shifting focus to §§ 102/103/112 (p. 24).
Broad Claiming: While breadth is not ineligibility, claims that recite mathematical calculations explicitly (like Example 47) rather than referencing a tool or machine (like Example 39) remain vulnerable (pp. 10, 11).
Conclusion
The USPTO’s recent updates signal a pragmatic approach to subject matter eligibility, particularly for AI. By clarifying that AI operations are not inherent “mental processes” and establishing that improvements to machine learning models are technological improvements, the Office is aligning with the current reality of software innovation (pp. 8, 30).
This shift places a renewed emphasis on the traditional patentability requirements. IP professionals should continue to ensure that specifications are robust enough to support technological improvements and that claims are structured to reflect specific tools and components rather than abstract results (p. 13).
Based on this webinar and the evolving USPTO guidance, practitioners may want to review in-draft software and AI patent applications to ensure the specification explicitly describes the technological improvement (e.g., speed, efficiency, storage reduction) and that claims recite the specific steps achieving that improvement, rather than just the (mathematical) result (p. 13).
Any clarification and (examiner) training is welcome, but for now it seems the Enfish/Desjardins playbook is to:
Detail the specific technical problem being solved (e.g., “catastrophic forgetting”).
Disclose the specific technical solution in the specification, including specific technological benefits.
Incorporate a claim limitation(s) that explicitly connect the abstract concepts to that tangible technological improvement.
Technology-based solutions to technology-based problems will continue to be leading the eligibility train, but determinations of what elements can be characterized as “conventional” are still too ambiguous and inconsistent.
Director Squires appears to be focused on reducing the administrative burden of §101. Recently, the USPTO issued memoranda regarding the use of subject-matter eligibility declarations (SMEDs).
Practitioners will continue to monitor developments on the Alice eligibility front.
Disclaimer: This is provided for informational purposes only and does not constitute legal or financial advice. To the extent there are any opinions in this article, they are the author’s alone and do not represent the beliefs of his firm or clients. The strategies expressed are purely speculation based on publicly available information. The information expressed is subject to change at any time and should be checked for completeness, accuracy and current applicability. For advice, consult a suitably licensed attorney and/or patent professional.