USPTO Director’s Panel Vacates AI Patent Rejection, Citing National Interest and Rebuking Overbroad § 101 Analysis
Ex parte Desjardins emphasizes technical improvement
In a significant patent office decision for artificial intelligence innovators, the Appeals Review Panel (ARP) of the USPTO, including new Director John A. Squires, has vacated a Patent Trial and Appeal Board (PTAB) panel’s patent eligibility rejection of an AI-related invention. The ruling serves as both a powerful endorsement for the patentability of AI-based innovations and a sharp rebuke of over-generalized § 101 analyses that ignore precedent regarding technological improvements.
The September 26, 2025 “Superseding Order Convening Appeals Review Panel and Granting Sua Sponte Rehearing” in Ex parte Desjardins (2024-000567) provides guidance on applying the § 101 framework to machine learning technologies, reinforcing that claims directed to technical improvements in computer functionality are patent-eligible, even when they involve mathematical concepts inherent in machine learning (Order, pp. 2, 8).
Background of the Dispute
The patent application at issue (No. 16/319,040) addresses a well-known technical challenge in machine learning known as “catastrophic forgetting” (p. 7). This problem occurs in continual learning systems, where a machine learning model trained to perform a first task (e.g., identifying cats) loses its proficiency after being subsequently trained on a second, different task (e.g., identifying dogs). The result is a model that performs well on the new task but has “forgotten” how to perform the original one.
The inventors, with DeepMind Technologies Limited as the real party in interest, proposed a computer-implemented method to solve this technical problem. Their method involves first determining an “importance” value for each parameter in the neural network relative to the first task. When the model is later trained on a second task, the training process uses an objective function that includes a penalty term based on these importance values (pp. 2-3).
This penalty discourages significant changes to parameters crucial for the first task, thereby “protecting performance” on the original task while still allowing the model to learn the new one. The specification states this approach reduces storage needs and system complexity by allowing a single model to learn multiple tasks sequentially (p. 3).
Procedurally, the application had a complex journey. After an examiner rejected the claims, a PTAB panel affirmed the rejection on § 103 obviousness grounds. However, that same panel went a step further, instituting its own new ground of rejection under 35 U.S.C. § 101, finding the claims directed to a patent-ineligible abstract idea.
After the panel denied rehearing, the case was elevated to the prestigious Appeals Review Panel, consisting of the USPTO Director, the Acting Commissioner for Patents, and the Vice Chief Administrative Patent Judge (p. 4).
The ARP’s Analysis
The central legal issue before the ARP was whether the claims, which recite a mathematical calculation, were directed to a patent-ineligible abstract idea under the framework established in Alice Corp. v. CLS Bank.
The ARP’s analysis focused squarely on the first step of the Alice test, which corresponds to Step 2A of the Manual of Patent Examining Procedure (MPEP) (p. 6).
The ARP first conceded, as the appellant did not dispute, that the claims recite a mathematical concept—specifically, “computing... an approximation of a posterior distribution”—which qualifies as an abstract idea under Prong One of the MPEP’s Step 2A analysis (pp. 6-7). The decision, therefore, turned entirely on Prong Two: whether the claim as a whole integrates that abstract idea into a practical application (p. 7).
Here, the ARP departed sharply from the original panel. The Board previously found “no additional element (or combination of elements) recited in Appellant’s claims [] that may have integrated the judicial exception into a practical application,” (p. 7) while the ARP found that the claims did, in fact, integrate the mathematical concept into a practical application that constituted a specific improvement to computer functionality.
The ARP said:
In particular, the Appellant identifies certain limitations of independent claim 1 and asserts that “the claimed subject matter provides technical improvements over conventional systems by addressing challenges in continual learning and model efficiency by reducing storage requirements and preserving task performance across sequential training,” citing paragraph 21 of the Specification for support. [citing Request for Rehearing] at 7-9; see also id. at 8 (”This training strategy allows the model to preserve performance on earlier tasks even as it learns new ones, directly addressing the technical problem of ‘catastrophic forgetting’ in continual learning systems.”). We agree with the Appellant. (p. 7)
The ARP connected the advantages detailed in the specification (e.g., reducing storage requirements and preserving task performance) to specific limitations in the claims.
When evaluating the claim as a whole, we discern at least the following limitation of independent claim 1 that 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.’ We are persuaded that constitutes an improvement to how the machine learning model itself operates, and not, for example, the identified mathematical calculation. (p. 9)
The ARP grounded its reasoning in the Federal Circuit’s seminal pro-eligibility decision in Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016), which held that claims are not abstract if they are directed to a specific improvement in the capabilities of a computer, rather than a generic application of an abstract concept (p. 8).
The ARP chastised the original PTAB panel for its “overbroad reasoning,” which it said “eschewed the clear teachings of Enfish“ (p. 9). The initial panel had improperly generalized the claims, equating machine learning with an unpatentable algorithm and dismissing the remaining elements as mere “generic computer components” (p. 9).
In a striking policy statement, the ARP warned of the broader consequences of such flawed analyses, stating, “Categorically excluding AI innovations from patent protection in the United States jeopardizes America’s leadership in this critical emerging technology” (p. 9).
Key Takeaways and Practical Implications
The Desjardins decision offers critical guidance for patent practitioners, inventors, and companies operating in the AI space.
Frame AI Inventions as Technical Solutions: The ARP’s focus on the improvement to the functioning of the machine learning model itself is instructive. Practitioners should draft applications and frame arguments to emphasize how the claimed invention provides a specific, concrete solution to a technical problem inherent in AI computer systems (e.g., catastrophic forgetting, memory optimization, processing efficiency), rather than merely applying a mathematical principle.
Focus on the Claim as a Whole: By evaluating the claim “as a whole” (p. 9), the panel sidestepped the conventional § 101 analysis that often involves dissecting a claim and dismissing elements as generic or well-known. Instead, the ARP’s approach suggests that if a claim, viewed holistically, is directed to a specific technological improvement, it can integrate an abstract idea into a patent-eligible practical application.
Connecting the Specification to the Claims is Crucial: While the claims ultimately define the legal scope of an invention, this decision reinforces how a well-drafted specification is essential for demonstrating patent eligibility. The ARP repeatedly looked to the specification’s description of the invention’s advantages to interpret the claims as a patent-eligible practical application (pp. 8-9). For practitioners, the lesson is that the technical challenges and their solutions, articulated in the specification, must be clearly linked to and reflected in the language of the claims. This synergy is critical for overcoming § 101 rejections.
Enfish Remains the North Star for Software Eligibility: The ARP’s strong reliance on Enfish signals to examiners and the PTAB that it remains the controlling precedent for determining the eligibility of software-based technological improvements. Arguments against § 101 rejections for AI/ML claims should continue to center on how the invention improves the computer’s functionality.
A Shift in Focus: Prioritizing Traditional Examination Tools: Perhaps the most direct guidance for future prosecution is the ARP’s pointed statement that §§ 102, 103, and 112 are the “traditional and appropriate tools” for examination (p. 10). This language strongly suggests a preference for using substantive prior art and written description rejections over the more abstract and often convoluted § 101 analysis. The procedural history of this case—where significant resources were expended to correct a sua sponte § 101 rejection while the underlying § 103 rejection remains—highlights the inefficiency the ARP may be seeking to curb.
Overall, the takeaway appears clear: while § 101 remains a hurdle, the USPTO leadership is signaling the core of patent examination should center on novelty, non-obviousness, and enablement, especially for AI-related inventions.
Perhaps the USPTO leadership believes that—because of the parallel analysis under sections 102 and 103—there is no need to explicitly discuss whether the finding of conventional or generic equipment is needed. Or, when there is a technical improvement, conventionality means less.
Last Thoughts (For Now)
The ARP’s decision in Ex parte Desjardins is a welcome and necessary course correction for applicants. It sends an unambiguous message from the highest levels of the USPTO that legitimate AI innovations providing tangible improvements to technology are patent-eligible.
By vacating the PTAB’s overreaching § 101 rejection and reaffirming the agency’s understanding of the case law, the decision hopefully provides greater clarity and predictability for innovators in one of the most important technological fields of today.
It encourages a more disciplined and precise application of § 101, pushing examiners and the Board to use the proper statutory tools to assess patentability without stifling progress in a critical area of American innovation.
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.