The recent expansion of patent elegibility for AI inventions before the USPTO

Introduction

The new United States Patent and Trademark Office (USPTO) Director John A. Squires was sworn in on September 22, 2025 and wasted no time that week in expanding patent eligibility for AI related inventions. In particular, the new Director presided over the September 26 Appeals Review Panel (ARP) decision in Ex parte Desjardins, Appeal 2024-000567. In its decision, the ARP begins explicitly steering USPTO claim interpretation policy under 35 U.S.C. § 101 in a new direction that aims to reduce patent eligibility scrutiny and potentially minimize the now-classic hurdles associated with interpreting abstract ideas and practical implementations thereof under the established Alice/Mayo framework.

In Desjardins, the ARP interpreted a claimed machine learning training pipeline as a technological improvement. In its analysis, the ARP identified at least the claim term of training a machine learning model including a plurality of parameters on a second machine learning task to “adjust first values of the plurality of parameters to optimize the machine learning model on the second machine learning task while protecting performance of the machine learning model on [a] first machine learning task” as constituting a patent eligible improvement on how the machine learning model operates. In supporting this position, the ARP referred to the specification-declared advantages of the claimed subject matter in terms of lower storage capacity requirements, reduced system complexity, and effectively learning new tasks without losing knowledge on previous tasks.

Until recently, such broadly claimed processing operations exemplified by the steps of adjusting parameter values and protecting past performance recited in Ex parte Desjardins were commonly interpreted as applying generic computer parts to an abstract idea. The ARP acknowledges its plausible departure from the previous norm practiced by most examiners and appeal panels, noting in its analysis that “under the [original] 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.” According to the ARP, “Examiners and panels should not evaluate claims at such a high level of generality.”

As put by Director Squires in a subsequent statement before the Subcommittee on Intellectual Property Committee on the Judiciary United States Senate October 9, 2025, “patent eligibility is not an abstract debate” but “a matter of national security, of resilience, and of ensuring that America’s system of innovation remains robust enough to confront the challenges of the twenty‑first century.” Advocating for less restrictive interpretation under 35 U.S.C. § 101, Director Squires’ statement further explains that “[s]ection 101 should not be misused as a blunt instrument to exclude entire technological fields” as “patent law must remain expansive if it is to remain true to its statutory text, to its history, and to its constitutional purpose.” In the few months since the ARP’s decision in Ex parte Desjardins, the Patent Trial and Appeal Board (PTAB) has been largely following Director Squires’ leadership, finding new acceptance for broadly drafted processing claims within the Alice/Mayo framework.

Patent Trial and Appeal Board Cases

In this regard, the PTAB panel in Ex parte Mittal, Appeal 2025-002097 (November 24, 2025) reversed patent eligibility rejections of a claimed method of retraining a deployed machine learning model to detect and correct data-drift over time. In its analysis, the PTAB identified the claimed method steps of generating “a validation dataset from live model predictions generating a validation dataset comprising a plurality of data points” in view of user preferences, “ranking the plurality of data points of the validation dataset in view of the user preferences”, and “retraining the deployed machine learning model utilizing a new training dataset based upon the validation dataset and the ranked plurality of data points” as reciting an improvement in the functioning of a computer rather than broadly directing use of a computer and machine learning.

Similar to the ARP in rehearing Ex parte Desjardins, the PTAB noted corresponding specification-declared advantages in automatically correcting data-drift and accounting for different parameters affected by user preferences. Furthermore, the PTAB directly cites Ex parte Desjardins as precedential, noting that “claims reciting particular improvements in training a machine-learning model reflected an improvement to technology.” With this precedent, the PTAB determined that the claimed method of retraining a deployed machine learning model in Ex parte Mittal recites a technological improvement in machine learning with sufficient specificity that distinguishes it from claims in other cases that were deemed abstract for merely applying machine learning or data visualization without disclosing any technology-specific method.

In another case, the PTAB panel in Ex parte Brush, Appeal 2025-002376 (November 17, 2025) reversed patent eligibility rejections of a claimed machine‑learning system that converts heterogeneous electronic health record data into model‑ready feature catalogs and iteratively improves model performance. In its analysis, the PTAB determined the claimed steps of “generating a prediction by running a first predictive model, of the one or more predictive models, against the set of features hosted in the first feature catalog, wherein the predictive model is configured to make predictions based on the correlations within the normalized population data; [and] evaluating the accuracy of the prediction by comparing the prediction to historical data; altering the first predictive model based on the accuracy of the prediction” integrate any mental process or abstract idea into a practical application.

Similar to Ex parte Desjardins and Ex parte Mittal, the PTAB in Ex parte Brush noted corresponding specification-declared advantages of the claimed machine‑learning system in addressing data transfer bottlenecks between data warehousing and analysis, enabling correlations within normalized data to drive predictions, facilitating model verification and updates as warehoused data changes, and avoiding bespoke, one‑off pipelines by using feature catalogs compatible across multiple predictive models. Furthermore, the PTAB directly cited to Ex parte Desjardins, noting its precedential weight in establishing that a “claim is patent eligible [when] it ‘reflects… an improvement to how the machine learning model itself operates.’”

In another case, the PTAB panel in Ex parte Wang, Appeal 2025-001388 (October 29, 2025) reversed patent eligibility rejections of a claimed machine learning pipeline that aligns multisensor time‑series data and trains a model to predict mechanical quality‑assurance failures. In its analysis, the PTAB identified the claimed steps of “training the self-learning application by submitting the modified corpus to the self-learning application,” including “using training data to perform the training,” “teaching the self-learning application to make a prediction of a likely failure… in response to the self-learning application identifying adverse conditions,” and “gaining experience, by the self-learning application, that allows the self-learning application to infer a semantic meaning from behavior of the set of attributes,” as not practically being performed in the human mind.

Similar to these other PTAB cases discussed above, the PTAB in Ex parte Wang noted corresponding specification-declared advantages in improvements to operations of a machine learning model. In this regard, the subject specification provided that time-aligned streams representing the time-series data culled from sensors in a manner allows the inference and correlation of various conditions and states of each attribute at different times and makes them appropriate for use as training data in a machine learning operation. Here, the PTAB again directly cites Ex parte Desjardins noting that the claimed steps and corresponding specification-declared advantages are similar to the “improvement to how the machine learning model itself operates” that the Board concluded “integrated the judicial exception into a practical application” in Ex parte Desjardins.

Notably, Ex parte Desjardins has not rendered any and all machine learning claims patent eligible. For example, the PTAB in Ex parte Kuusela, Appeal 2025-001619 (November 24, 2025) affirmed patent eligibility rejections of a claimed method of radiology therapy planning that lacked any limitations directed toward modifying or developing a machine learning model. In this regard, the claim at issue merely recites a computer-implemented method including accessing patient information for a patient, accessing an integrated dose prediction model that integrates a plurality of predictive models, selecting one or more predictive models, processing said patient information, and outputting the radiation dose distribution, with no additional elements that affect the form or function of the integrated dose prediction model. Here, the PTAB again directly cites Ex parte Desjardins in noting that the claimed method merely applies a judicial exception using generic computer components and does not improve the functioning of the computer itself, and lacks any improvement to computer functionality or to how the machine learning model itself operates.

Implications

Taken as a whole Ex parte Mittal, Ex parte Brush, and Ex parte Wang strongly indicate the PTAB is clearly following the precedent established in Ex parte Desjardins, which embody Director Squires’ statement before the Subcommittee on Intellectual Property Committee on the Judiciary United States Senate calling for an expanded interpretation of patent eligible subject matter. More pointedly, claim limitations reciting retraining deployed models as in Ex parte Mittal, converting health record data into model‑ready feature catalogs as in Ex parte Brush, or structuring machine learning pipelines as in Ex parte Wang may well have found difficulty in establishing patent eligibility without new guidance from the Ex parte Desjardins decision. It should be noted that each of these cases includes the PTAB reading advantages from the patent specification, and understood that one strategic approach to better position AI inventions for eligibility is to clearly pair claimed machine learning structures to explicit nuanced advantages within the patent specification.

In this manner, the PTAB has started to set a pattern of decisions showing expanded avenues of patent eligibility for unique machine learning models that may require little more structure than the method of radiology therapy planning provided in Ex parte Kuusela. While this article focused directly on machine learning structures, other recent decisions by the PTAB following Ex parte Desjardins with subject matter outside the immediate scope of this article have applied similar reasoning to system processing claims. Examples of such claims include those employing concurrent processing in Ex parte Williams, Appeal 2025-001079 (October 30, 2025); employing an AI model to change data stream formats based on detected circumstances in Ex parte Goyal, Appeal 2025-001692 (November 24, 2025); and employing application search processing that uses tracked user interaction signals from a first application to estimate intent and to modify result ranking delivered by a second application in Ex parte Paris, Appeal 2025-001701.

Notably, this new policy shift at the PTAB is occurring entirely within the existing statutory and regulatory framework, without requiring Congressional amendment or rulemaking, including any action on the proposed Patent Eligibility Restoration Act of 2025. The Federal Courts have likewise not yet addressed these emerging eligibility approaches, and it remains to be seen whether the Courts will adopt the same interpretive posture. For now, the PTAB decisions following Ex parte Desjardins signal a meaningful recalibration of patent eligibility analysis at the USPTO that should materially influence drafting and prosecution strategy moving forward unless and until the Courts or Congress intervene.