Sign Up for PublicationsEx Parte Desjardins: Paradigm Shift in AI Eligibility Under § 101
On September 26, 2025, the Appeals Review Panel of the United States Patent and Trademark Office (“USPTO”) issued a precedential decision in Ex parte Desjardins, Appeal No. 2024-000567. This decision has already begun reshaping patent eligibility analysis for artificial intelligence and machine learning inventions. Authored by newly confirmed USPTO Director John A. Squires, the decision vacated a previous appellate rejection under 35 U.S.C. § 101, holding that claims directed to training machine learning models can be patent eligible under certain circumstances. Designated as precedential on November 4, 2025, and incorporated into updates to the Manual of Patent Examining Procedure (“MPEP”) on December 5, 2025, Desjardins has already become a seminal decision in § 101 jurisprudence and will carry significant influence for years to come.
The patent application at issue in Desjardins relates to methods for training machine learning models on multiple tasks sequentially. The claimed invention addresses the technical problem of “catastrophic forgetting”—the tendency of neural networks to lose knowledge of previously learned tasks when trained on new ones. The specification identifies concrete advantages to the claimed method, including reduced storage requirements and system complexity, and the ability to learn tasks in succession while protecting prior knowledge.
During prosecution, the examiner rejected the claims under § 103 for obviousness but did not raise a § 101 issue. On appeal, in March 2025, the Patent Trial and Appeal Board (“PTAB”) affirmed the § 103 rejection and, acting sua sponte, entered a new ground of rejection under § 101, finding the claims directed to an unpatentable “mathematical algorithm” with only “generic computer components” as additional elements. Rehearing was denied in July 2025, with the panel citing the Federal Circuit’s decision in Recentive Analytics, Inc. v. Fox Corp., 134 F.4th 1205 (Fed. Cir. 2025). In August 2025, an Appeals Review Panel (“ARP”) was convened sua sponte to review both decisions.
Applying the two-step Alice framework, the ARP agreed that the claims recited at least one abstract idea—a mathematical calculation involving computation of a posterior distribution approximation. But, critically, and citing Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016), the panel found that the claims reflected concrete improvements to how the machine learning model itself operates. Specifically, the ARP identified the limitation requiring the model to “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” as constituting an improvement to the technology, not merely a mathematical calculation applied in a generic computing environment.
Newly-minted Director Squires used the opinion to deliver pointed and precedential guidance. The decision cautioned that “[c]ategorically excluding AI innovations from patent protection in the United States jeopardizes America’s leadership in this critical emerging technology.” The ARP criticized the lower panel for evaluating claims “at such a high level of generality” that it “essentially equated any machine learning with an unpatentable ‘algorithm’ and the remaining additional elements as ‘generic computer components,’ without adequate explanation.”
Further, Director Squires signaled a reorientation of examination priorities, stating: “This case demonstrates that §§ 102, 103 and 112 are the traditional and appropriate tools to limit patent protection to its proper scope. These statutory provisions should be the focus of examination.” This language strongly suggests that the USPTO sees § 101 as a threshold inquiry, with the substantive work of limiting patent scope performed by novelty, nonobviousness, and written description requirements.
Desjardins provides clear guidance for practitioners with pending AI patent applications in their portfolio. Applications should consider identifying a concrete technical problem and explain in detail how the invention provides a technical solution within the specification. Practitioners should consider framing AI innovations as improvements to the functioning of the model or system itself, rather than applications of generic techniques to new domains. And given the MPEP updates incorporating Desjardins, practitioners should cite the decision directly when responding to § 101 rejections. Indeed, Desjardins has already impacted subsequent decisions: in Ex parte Carmody, No. 2025-002843 (PTAB, Dec. 30, 2025), the PTAB reversed a § 101 rejection where claims recited a modular machine learning architecture with defined training datasets that improved how the system operated—reasoning closely tracking the Desjardins framework.
Desjardins represents a significant shift in the USPTO’s approach to AI inventions. It establishes that purely softwarebased improvements to AI models can be patent-eligible and clarified that examiners should rely on §§ 102, 103, and 112 rather than § 101 as the primary tools for limiting patent scope. Although questions remain about how courts will harmonize this guidance with the Federal Circuit’s Recentive line of authority, Desjardins provides the strongest foundation in years for securing patent protection for AI innovations. Its rapid incorporation into the MPEP, precedential designation, and immediate influence on subsequent PTAB decisions underscore that the USPTO has shifted its guidance on and approach to AI eligibility under § 101.