More Than Math: How Desjardins Recognizes AI Innovations as Patent-Eligible Technology

On November 4, 2025, the USPTO in Ex parte Desjardins designated as precedential an earlier Appeals Review Panel (ARP) decision overruling the Patent Trial and Appeal Board (Board), instead holding that claims directed to training a machine learning model are patent-eligible under 35 U.S.C. § 101 when they integrate a mathematical concept into a practical application that improves how the model operates. That precedential designation represents a material shift in legal precedent in the § 101 arena post-Alice, and it has already driven updates to the MPEP and examiner practice. As a result, Desjardins signals an adjustment in practice in favor of AI and software eligibility at the USPTO.

What Desjardins Decided

The claims in Desjardins recite a computer-implemented method of training a machine learning model on a first task, determining importance measures for parameters by computing an approximation of a posterior distribution, and then training the model on a second task while adjusting 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 . . . .”

Although both the ARP and the Board agreed that the mathematical calculation recites an abstract idea, the ARP disagreed with the Board’s conclusion that no additional elements integrate the judicial exception into a practical application. Relying on Enfish, which recognizes that “Software can make non-abstract improvements to computer technology just as hardware improvements can,” the ARP found that the specification identified technological improvements including the ability to “effectively learn new tasks in succession whilst protecting knowledge about previo.tasks,” thus allowing AI systems to “us[e] less of their storage capacity” and enabling “reduced system complexity.” With that in mind, the ARP concluded that the limitation requiring adjustment of parameters to optimize performance on a second task while protecting performance on the first task reflected an “improvement to how the machine learning model itself operates, and not, for example, the identified mathematical calculation.” The ARP nevertheless held that the claims remain rejected under § 103, demonstrating that “§§ 102, 103 and 112 are the traditional and appropriate tools to limit patent protection to its proper scope.”

What the USPTO Has Done Since

Having decided that Desjardins recites an “improvement to how the machine learning model itself operates,” the ARP turned to warning of the consequences of deciding otherwise in the future, namely, that “Categorically excluding AI innovations from patent protection in the United States jeopardizes America’s leadership in this critical emerging technology.”

The ARP criticized rejections that “essentially equated any machine learning with an unpatentable ‘algorithm’ and the remaining additional elements as ‘generic computer components,’ without adequate explanation.”

The USPTO has moved quickly to implement Desjardins. On November 4, 2025, the USPTO designated the decision as precedential. On December 5, 2025, the USPTO issued updates to the MPEP instructing examiners to evaluate the specification for a detailed, non-conclusory disclosure of a technological improvement and confirm that the claim “as a whole” reflects that improvement. Examiners must avoid overgeneralizing or dismissing limitations as “generic computer components” when they confer technical improvement. The MPEP now treats improved machine learning training that preserves prior knowledge as an example of improvement in computer functionality.

Why This Matters

Desjardins establishes a practical path for AI and software eligibility. Generally, according to the ARP, if the specification describes and the claim reflects a technical improvement, eligibility should be resolved positively. The USPTO is now crediting problem-oriented disclosures and claim language that solves specific technical challenges as evidence of practical application and improvement.

Practitioners should expect more consistent credit for technical improvements in AI and software. At the same time, examiners will likely scrutinize whether claims truly reflect the disclosed improvement and whether the specification’s assertions are sufficiently detailed rather than conclusory.

As the ARP made clear, the debate is shifting to §§ 102, 103, and 112, which will likely require practitioners to build a solid record to overcome these traditional hurdles from the start.

Key Takeaways

• Expect increased scrutiny under §§ 102, 103, and 112 as § 101 rejections become less common for well-drafted AI and software applications.
• U.S. patent attorneys should consider treating Desjardins as the operative playbook for AI and software eligibility: (1) identify a technical problem, (2) disclose a technical solution/improvement in detail, and (3) draft claims that implement that solution/improvement.
• U.S. patent attorneys should broaden their claim formats to cover eligible types of AI model improvements and corresponding patent eligible subject matter, and should take additional care to ensure that patent applications intended to be filed overseas still provide express support for specific technical applications and architectures when applicable.
• Non-U.S. patent attorneys and agents drafting priority-establishing patent applications outside the U.S. should keep in mind that the U.S. currently permits a broader scope of protection than perhaps their domestic territories permit and should consider including appropriate claim formats for use in the U.S.
• An improvement to an AI model per se is not likely to be deemed patentable subject matter at the EPO unless it is specifically tied to a type of technical effect deemed acceptable in EPO case law. Typically, this requires the model to be limited to an application in a technical field, or, alternatively, the model would need to be adapted to a particular technical implementation. We recommend seeking advice on the specifics from a suitably specialized patent attorney.

A Note on the ARP and Precedential Designation

On July 24, 2023, the USPTO established the Appeals Review Panel (ARP). The ARP may be convened by the Director sua sponte to review decisions of the Board in ex parte appeals, re-examination appeals, and reissue appeals. Requests for ARP review will not be considered.

ARP review decisions are routine decisions. But, ARP’s routine decisions may be nominated after issuance for precedential designation. The Director ultimately decides whether to designate a decision as precedential. A precedential decision is binding Board authority in subsequent matters involving similar facts or issues.

Here, the Director’s designation of Desjardins as precedential means that examiners must follow its reasoning in future cases.