Artificial intelligence (“AI”)-related patents sit at the center of fast-evolving policy and doctrinal shifts at the United States Patent and Trademark Office (“USPTO”). Under Director John A. Squires, the USPTO has begun to recalibrate how 35 U.S.C. § 101 (“Section 101”), which defines what subject matter is patent-eligible under US law, is applied to AI inventions. Section 101 has frequently been used by patent examiners at the USPTO to find AI inventions unpatentable for being nothing more than abstract ideas. Since Director Squires’ confirmation in mid-September 2025, the USPTO has begun to take into account concrete technological improvements found in AI patent applications and redirect examination intensity toward 35 U.S.C. § 102 (novelty), 103 (obviousness), and 112 (written description/enablement).

The recalibration by the USPTO towards an expansive view of patent eligibility is exemplified by a recent Appeals Review Panel (“ARP”) decision in Ex parte Desjardins. Director Squires convened the APR to review the USPTO’s Patent Trials and Appeals Board (“PTAB”) decision regarding the patentability of US Patent Application No. 16/319,040. The USTPO recalibration in AI patentability cases has been further reinforced by the Director’s public remarks and early administrative actions. At the same time, the Federal Circuit’s recent guidance in AI cases remains comparatively stringent, setting up an important tension to watch.

**The ARP’s Desjardins Decision: From Algorithm-as-Abstract to Improvement-as-Eligible  **

Only two days after being sworn in, Director John Squires convened an Appeals Review Panel (“ARP”) to review the Patent Trials and Appeals Board (“PTAB”) decision in Ex parte Desjardins.

• The Ex parte Desjardins case involved a Google/DeepMind patent application aimed at mitigating “catastrophic forgetting,” a well-known machine learning challenge in sequential task learning.
• The PTAB panel had affirmed the Examiner’s rejection of the claims as obvious. The panel further added a new Section 101 rejection, characterizing the Ex parte Desjardins claims as directed to a “mathematical algorithm” implemented on generic hardware and thus as unpatentable subject matter.
• The ARP vacated the Section 101 subject matter eligibility rejection and left the obviousness rejection intact.  
• The ARP’s Ex parte Desjardins decision accepts that the claims at issue implicate an abstract idea; however, the ARP’s decision found the claims at issue to be patent eligible because they are “directed to an improvement to how the machine learning model itself operates.
• The ARP noted the claimed invention reduced memory usage, decreased system complexity, and preserved prior knowledge while learning new tasks.  The ARP reiterated that software claims can be eligible where they improve computer functionality or another technical field, even when improvement is expressed through logical structures and processes rather than novel hardware.  
• The ARP further critiqued the PTAB’s analysis as “overbroad,” cautioning against equating “any machine learning” with an unpatentable algorithm or hand-waving real technical elements as generic components.

 

What is clear from the ARP’s Ex parte Desjardins decision is the following: the Director considers 35 U.S.C. § 102, 103, and 112 to be the “traditional and appropriate tools” to limit patent scope, and USPTO examiners should avoid high-level generalities that flatten technical detail, particularly in claims related to AI-based inventions. The ARP’s decision explicitly refocuses subject matter eligibility doctrine away from categorical exclusions and back toward a technology-specific, improvement-oriented inquiry.

**Policy Signals from the Director: Eligibility as Innovation and Security Policy  **

The ARP decision is buttressed by a broader suite of early actions and public statements by Director Squires and the USPTO.

• Within days of taking office, Director Squires presided over a ceremony issuing patents in two technology areas that have historically faced heavy Section 101 headwinds—medical diagnostics and distributed ledger technology—using the moment to connect applied technologies to the historical arc of American patent law.
• Director Squires’ recent October 9, 2025 Senate statement set forth his view that an expansive view of patentable subject-matter is not only faithful to the statutory text and precedent but also a national security imperative. In his statement, he also warned against using Section 101 as a blunt instrument that deters investment and cedes technological leadership abroad.
• The USPTO release of the August 9, 2025 Memorandum to the Examining Corp, which indicated that rejections under Section 101 require a determination that a claim is "more likely than not" ineligible. The Memorandum further reminded the Examining Corps that “uncertainty is not enough for a rejection” and that care should be taken to not “over-simplify” claims when reviewing the technical improvements to the software or machine learning model used in the claimed invention.
  Collectively, these acts and statements form a clear policy vector: treat AI and software inventions as potentially eligible when they improve technical systems; avoid categorical bars; and return the heavy lifting of narrowing to novelty, non-obviousness, and disclosure doctrines.

**The Federal Circuit’s Recentive Decision: A Stricter Backdrop  **

While recent USPTO decisions may be heartening for applicants seeking to protect their AI-based inventions with broader patents, Applicants need to consider recent Federal Circuit cases.

• In Recentive Analytics, Inc. v. Fox Corp., the Federal Circuit set a high bar when assessing patent eligibility of AI claims. The court in Recentive held that claimed inventions that merely apply established machine learning methods to a new data environment, without disclosing specific improvements to the machine-learning model or computer functionality, are not patent eligible.
• The PTAB invoked Recentive Analytics in its Ex parte Desjardins decision, noting that iterative training “is incident to the very nature of machine learning” and does not itself rescue claims from abstraction.

The ARP acknowledged Recentive but distinguished Ex parte Desjardins on technical grounds. Per the ARP, the claimed training approach at issue in Ex parte Desjardins purportedly improved how the model operates—conserving memory, reducing complexity, and preserving learned knowledge—rather than merely transposing generic ML techniques to a new dataset.

An essential question for AI invention patent eligibility going forward is this: do the claims recite concrete improvements to model operation or computer functionality, grounded in the specification, or do they simply deploy known ML routines to a different problem domain?  Where the former is credibly shown, USPTO examination is poised to be receptive; where the latter is all that is offered, Federal Circuit precedent may still be fatal in litigation or PTAB review. 

**Practical Implications for Applicants **

• Applicants should anchor patent eligibility positions for AI inventions in specific technical improvements described in the specification—especially quantifiable effects like memory savings, reduced training complexity, or enhanced performance stability—and connect those improvements to claim language.  
• Applicants should vigorously challenge Section 101 rejections where an office action rejection lacks specificity or disregards technical details of the claimed invention which provide concrete improvements to model operation or computer functionality.
• 35 U.S.C. § 102, 103, and 112 will likely be relied upon more by USPTO Examiners, and applicants should be prepared for more rigorous prior-art and disclosure scrutiny.

 

**Conclusion  **

The USPTO appears to be becoming more welcoming to AI-related eligibility, focused on technical improvements within machine learning models and training approaches, and resistant to categorical disqualification of AI claims. The open question is will the courts follow the USPTO’s lead? AI protection by patent law in the U.S. is at a pivotal moment. The ARP’s Ex parte Desjardins decision reframes how USPTO Examiners will analyze AI inventions under 35 U.S.C. 101. For innovators in AI, the near-term opportunity is real—but so is the imperative to build eligibility on a bedrock of specific, demonstrable improvements.