TAKEAWAY: Together, Desjardins and the Memorandum mark a clear shift towards recognizing technical improvements in AI and machine-learning inventions as patent-eligible under 35 U.S.C. §101.
In August 2025, the USPTO released a memorandum (“Memorandum”) pertaining to the subject-matter eligibility of artificial intelligence (“AI”) and machine-learning inventions under 35 U.S.C. §101. As background, in 2019 the USPTO began requiring that any eligibility rejections fall into three categories: mathematical concepts, certain methods of organizing human activity, and mental processes. The eligibility test further analyzes whether the claims recite a judicial exception (Step 2A, prong one), and whether they are, as a whole, integrated into the practical application (Step 2A, prong two). The Memorandum clarifies the “mental process” category, emphasizing that claims that are not practically performable in the mind do not recite a judicial exception (i.e., passes prong one). Further, if a claim merely involves a judicial exception, but does not recite it, the claim also would clear Step 2A muster.
Regarding prong two, the Memorandum emphasizes that examiners must analyze claims as a whole and not isolate individual computational elements. Furthermore, the Memorandum stresses that claims should not be rejected unless it is more likely than not that they fail eligibility. In short, the Memorandum aims to reduce overbroad rejections and promote a more practical application of Step 2A, ensuring that meaningful technological improvements are recognized.
Just weeks later, the Appeals Review Panel’s (“ARP”) rehearing decision in Ex parte Guillaume Desjardins, Decision on Request for Rehearing, No. 2024-000567 (USPTO Admin. Rev. Panel Sept. 26, 2025) (“Desjardins”) applied those very principles in practice. The subject patent application claimed a machine-learning method for sequentially training models while preserving previously learned knowledge, an approach to prevent catastrophic forgetting of model parameters learned from prior tasks. The PTAB panel had introduced a § 101 rejection, characterizing the invention as an abstract mathematical computation. On rehearing, the ARP vacated that rejection, holding that the claimed steps, particularly the selective parameter adjustment for optimizing performance of the machine learning model on a second learning task while protecting its performance on a first learning task, reflected a technical improvement in model performance. The ARP emphasized that eligibility must hinge on whether the claim integrates the abstract idea into a practical application, rather than on whether it happens to use equations or algorithms.
Viewed together, the Memorandum and Desjardins reflect a coherent internal shift at the USPTO towards greater nuance and restraint in AI patent eligibility analysis. The Memorandum provides the procedural and interpretive foundation, cautioning examiners against overbroad ‘mental process’ characterizations and instructing them to evaluate the claim’s full interplay of elements. The ARP’s decision then provides the application of those principles, demonstrating how machine-learning innovations that enhance model operation can qualify as patent-eligible under Step 2A of the Alice/Mayo framework. This alignment signals how the Office is seeking a balanced path, one that acknowledges that AI inventions are not abstract by default and that eligibility must focus on the claimed technical contribution. This intended alignment is further evidenced by the ARP’s statement in its decision from Desjardins that “[c]ategorically excluding AI innovations from patent protection in the United States jeopardizes America’s leadership in this critical emerging technology.”
For patent practitioners, these developments offer concrete tools for arguing the eligibility of AI and machine-learning inventions. When facing a §101 rejection, an Applicant may consider arguing how the claims describe technical improvements in model operation, data handling, or computational efficiency, based on the guidance recently provided by the Memorandum and Desjardins. Pointing examiners to the Memorandum’s instructions to avoid overgeneralizing algorithmic steps as ‘mental processes,’ and citing to Desjardins as evidence that improvements to machine-learning models can constitute a “practical application,” may be resourceful approaches during prosecution when seeking to demonstrate patent eligibility under 35 U.S.C. §101.