by Dennis Crouch
In a striking early display of directorial authority, newly sworn-in Under Secretary of Commerce and USPTO Director John Squires has vacated a Patent Trial and Appeal Board (PTAB) decision that had entered a new ground of rejection under 35 U.S.C. § 101 against DeepMind Technologies’ machine learning patent application. The September 26, 2025 decision in Ex parte Desjardins, Appeal 2024-000567, Application 16/319,040, represents one of Director Squires’ first substantive actions since taking office one week ago and offers yet another clear signal that Section 101 reform is building steam. SquiresPTABDecision.
Squires writes:
Categorically excluding AI innovations from patent protection in the United States jeopardizes America’s leadership in this critical emerging technology. . . the panel essentially equated any machine learning with an unpatentable ‘algorithm’ and the remaining additional elements as ‘generic computer components,’ without adequate explanation. . . . most troubling [the Panel] eschewed the clear teachings of Enfish, and instead substituted only a cursory analysis that ignored this well-settled precedent. Panels should treat such precedent with more care, especially when acting sua sponte.
Finally, and most clearly a signal to all USPTO employees to back-off with regard to patent eligibility, he writes:
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.
The application, filed by DeepMind (owned by Google) and prosecuted by Fish & Richardson, claims methods for training machine learning models on multiple sequential tasks while avoiding “catastrophic forgetting,” which is defined as the phenomenon where neural networks lose knowledge of previous tasks when trained on new ones. As someone with some ADHD tendencies, I recognizing this issue.
Independent claim 1 recites computing “an approximation of a posterior distribution over possible values of the plurality of parameters” and training the model 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.” The specification points to technical improvements including reduced storage requirements and preserved task performance across sequential training sessions.
A PTAB panel initially affirmed the examiner’s rejection of all pending claims under 35 U.S.C. § 103 and entered a new ground of rejection under § 101, determining the claims were directed to an abstract idea without an inventive concept. DeepMind filed a request for rehearing, which the Board denied on July 14, 2025. Before Director Squires took office, an Appeals Review Panel had been convened to conduct sua sponte review of the Board’s decisions, focusing particularly on the § 101 rejection. The ARP, consisting of Director Squires, Acting Commissioner for Patents Valencia Martin Wallace, and Vice Chief Administrative Patent Judge Michael W. Kim, issued its decision on September 26, 2025, vacating the § 101 rejection while leaving undisturbed the Board’s affirmance of the § 103 rejection.
Patent Eligibility Under Section 101 35 U.S.C. § 101 provides that "any new and useful process, machine, manufacture, or composition of matter." Although this language has remained virtually unchanged in the statute for 200+ years, the Supreme Court greatly expanded its import in Alice Corp. v. CLS Bank Int'l, 573 U.S. 208 (2014) -- establishing a broad two step inquiry into whether a patent is improperly "directed to" a patent-ineligible concept.
The Appeals Review Panel’s analysis turned on Alice Step 2—whether the claims integrated the abstract idea (a mathematical calculation) into a practical application. The original Board panel had concluded that DeepMind’s claims recited mathematical concepts and lacked additional elements integrating the exception into a practical application. The ARP disagreed, finding that the claims “reflect an improvement to how the machine learning model itself operates.” Director Squires’ opinion emphasized that the claimed 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” constitutes “an improvement to how the machine learning model itself operates, and not, for example, the identified mathematical calculation.”
The decision relies heavily on Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016), which recognized that software innovations can constitute non-abstract improvements to computer technology even when defined by logical structures rather than physical features. The ARP determined that improvements identified in paragraph 21 of the specification—including the ability to “effectively learn new tasks in succession whilst protecting knowledge about previous tasks,” use less storage capacity, and enable reduced system complexity—were reflected in the claim language itself, distinguishing the case from situations where specification assertions alone prove insufficient.
Director Squires’ opinion contains remarkably direct criticism of the original Board panel’s reasoning and procedural choices. The decision states that “[u]nder a charitable view, the overbroad reasoning of the original panel below is perhaps understandable given the confusing nature of existing § 101 jurisprudence, but troubling, because this case highlights what is at stake.”
Notably, the ARP did not disturb the Board’s affirmance of the § 103 rejection, and the opinion notes that “the claims at issue stand rejected under § 103,” limiting the practical benefit to DeepMind despite the § 101 victory. Interestingly, one PTAB judge had issued a concurring opinion that was opposite to Dir. Squires — arguing that the claims were ineligible, but that the obviousness rejection should be reversed.
For patent prosecutors handling AI and machine learning applications, the decision provides both encouragement and a roadmap.