AI and Software Patents in 2025: New Leadership and § 101 Eligibility Guidance

Top Takeaways from 2025

• Limit “mental processes” to what humans can actually perform. Only treat a step as a mental process if it can practically be performed in the human mind.
• Separate “recites” from “involves” abstract ideas. Many AI claims merely involve mathematical concepts rather than reciting a judicial exception, and in those cases, a Step 2A/2B analysis may not even be necessary.
• Anchor eligibility in concrete technological improvements. Focus on specific improvements to computer functioning or a technical field, not just automation, efficiency gains, or applying standard machine learning (ML) to new data.
• Analyze the claim as a whole. Determine whether the claim merely says “apply it on a computer/neural network” or instead claims a particular technical solution or improved way of achieving a result.
• Provide evidence in close cases using subject matter eligibility declarations (SMEDs) and the preponderance standard. In borderline situations, remember the “more likely than not” standard and consider using SMEDs to document mental-process limits or technological improvements.

The United States Patent and Trademark Office (USPTO) has recently undergone significant changes both in leadership and in Section 101 subject matter eligibility guidance. John Squires has been appointed as undersecretary of commerce for intellectual property and director of the USPTO. Under his leadership, the USPTO has issued updated eligibility guidance that builds on and refines its 2024 AI-focused direction and has influenced PTAB practice through key Appeals Review Panel decisions such as Ex parte Desjardins.¹ 

The 2024 guidance, published on July 17, 2024, introduced the AI-SME Update, which set out foundational principles for evaluating AI-related patent applications under 35 U.S.C. § 101. Among other things, it emphasized the distinction between claims that “recite” a judicial exception and those that merely “involve” one. Claims recite a judicial exception when they explicitly set forth or describe an abstract idea, such as by naming specific algorithms like backpropagation or gradient descent. By contrast, claims that merely involve an exception are based on or utilize an abstract idea without explicitly reciting it in the claim language. The 2024 guidance also made clear that simply incorporating AI, large language models, or neural networks into an abstract idea does not transform that idea into a patent-eligible application, and it included Examples 47-49 to illustrate the eligibility analysis for AI-related claims.

Against this backdrop, on April 18, 2025, the Federal Circuit’s decision in Recentive Analytics v. Fox Corp.² serves as a cautionary counterpoint that helped define the 2024 guidance and highlights what to avoid in AI claiming. The court held that patents applying ML models to generate optimized network maps and event schedules were directed to abstract ideas and ineligible under Section 101 because they merely applied conventional ML techniques to a new data environment, without any improvement to the ML technology or computer functioning itself. This outcome underscores that patent applications, beyond reciting functional results, must set forth concrete implementations that solve technological problems and improve computer or ML capabilities.

On August 4, 2025, the USPTO issued a Section 101 subject matter eligibility memorandum that, among other things, clarifies how examiners should analyze AI-related claims and offers a series of reminders and clarifications that, in practice, tend to favor applicants. First, it reiterates that examiners should not expand the “mental process” grouping to encompass claim limitations that cannot practically be performed in the human mind, and it expressly recognizes that AI-related limitations that cannot realistically be performed mentally fall outside this grouping, a critical constraint on overbroad Section 101 rejections. 

Second, it strengthens the 2024 “recites” versus “involves” distinction by providing concrete examples, distinguishing claims that merely involve mathematical concepts (such as “training a neural network”) from claims that actually recite judicial exceptions by naming specific algorithms like backpropagation or gradient descent. 

Third, the memo stresses that Step 2A, Prong Two analysis must assess how additional elements work together with any recited exception, rather than evaluating limitations in isolation. Examiners are instructed to adopt a holistic, claim-as-a-whole approach that considers how limitations interact and influence one another in determining whether an exception is integrated into a practical application. 
Fourth, the guidance offers more detailed instruction on separating true technological improvements from mere automation, requiring examiners to evaluate whether claims present particular solutions to problems or particular ways to achieve outcomes, rather than just the idea of a solution, and warning against oversimplifying claims under the “apply it” rubric; the memo recognizes that the “apply it” and improvement inquiries often overlap. 

Finally, the memorandum highlights a significant practical point: examiners should make a Section 101 rejection only when it is more likely than not (greater than 50%) that the claim is ineligible, explicitly anchoring the analysis in a preponderance-of-the-evidence standard and discouraging rejections in close cases.

On September 26, 2025, the USPTO issued its Ex parte Desjardins Appeals Review Panel decision, which further clarifies the contours of eligibility for AI-related inventions. In that decision, issued by Director Squires, the office overturned a Section 101 rejection and held that improvements to the functioning of machine learning models can constitute practical applications under the subject matter eligibility framework. Desjardins signaled an increased willingness to treat ML model improvements as technological improvements and to shift the focus of examination toward Sections 102 and 103, rather than reflexively relying on subject matter eligibility rejections.

On December 4, 2025, the office introduced clarifying guidance on SMEDs under 37 C.F.R. § 1.132, giving applicants a voluntary mechanism to submit factual evidence relevant to eligibility determinations. The two memoranda explain that applicants may use declarations to provide evidence such as proof of technological improvement, the state of the art at the time of filing, or information showing how a judicial exception is integrated into a practical application. Examiners must evaluate properly submitted SMEDs as part of the record under the same preponderance-of-the-evidence standard. The guidance emphasizes that SMEDs are most effective when presented in a declaration focused solely on subject matter eligibility, supported by objective evidence tied to the claimed invention, and that they are not a vehicle to supplement the original disclosure.

Taken together, the developments in 2025 built on the foundation laid by the 2024 AI-SME guidance, layering in a new director, fresh Federal Circuit authority, and a precedential PTAB decision to create a more coherent Section 101 landscape for AI and software. Under Director Squires, the USPTO’s August 4 memorandum, the Federal Circuit’s decision in Recentive, the ARP’s ruling in Desjardins, and the new SMED guidance all point in the same direction: claims that emphasize concrete technological improvements, are framed as more than mental processes or generic automation, and are supported by a clear evidentiary record stand a far better chance of clearing the eligibility hurdle.

---

¹Ex parte Desjardins, Appeal 2024-000567 (decided September 26, 2025).
²Recentive Analytics, Inc. v. Fox Corp., 1:22cv1545.