USPTO Codifies Desjardins: New MPEP Instructions for AI Eligibility
Let's Get Technical
On December 5, 2025, Deputy Commissioner for Patents Charles Kim issued a memorandum to the Patent Examining Corps regarding “Advance notice of change to the MPEP in light of Ex Parte Desjardins.” This guidance follows the designation of the much-discussed Appeals Review Panel (ARP) case, Ex Parte Desjardins, Appeal No. 2024-000567, as a precedential decision on November 4, 2025 (p. 1).
The memorandum instructs examiners on applying the Desjardins holding to eligibility determinations under 35 U.S.C. § 101, specifically revisions to MPEP §§ 2106.04(d), 2106.05(a), and 2106.05(f).
For the patent bar, these updates reinforce an emphasis under Director Squires for the USPTO to recognize specific technical improvements in machine learning training methods as patent-eligible subject matter, provided the specification offers sufficient technical detail.
The Desjardins Precedent: Solving “Catastrophic Forgetting”
By now, practitioners are likely well aware of this ARP opinion. This memo, however, highlights a few core concepts that the examiners need to take away. Most notably, the guidance draws attention to the ARP’s specific findings regarding “catastrophic forgetting,” where the panel “credited benefits including reduced storage, reduced system complexity and streamlining, and preservation of performance attributes associated with earlier tasks during subsequent computational tasks” (p. 2).
The memo underscores that because the specification disclosed these technological improvements, the claims integrated the abstract idea into a practical application at Step 2A, Prong Two (p. 2).
To reinforce this, the text explicitly reminds examiners of the Enfish standard: “software can make non-abstract improvements to computer technology, just as hardware improvements can” (p. 1).
MPEP Revisions: A Review
The memorandum implements several specific changes to the MPEP, effective immediately. These revisions place a heavy burden on the specification to support eligibility arguments.
MPEP § 2106.04(d) – Identifying the Improvement
A new subsection III has been added to MPEP § 2106.04(d) to summarize Desjardins. The text explicitly states that the claims were eligible because they “reflected the improvement disclosed in the specification” (p. 2).
The key takeaway for drafters is the nexus between the claim language and the technical benefits.
The ARP found that the limitation to “adjust the first values of the plurality of parameters to optimize performance... while protecting performance... on the first machine learning task” was not merely a mathematical result but a reflection of the “improvement disclosed in the specification” (p. 2).
MPEP § 2106.04(d)(1) – The “Conclusory” Warning
Perhaps the most significant instruction for practitioners appears in the revised second paragraph of MPEP § 2106.04(d)(1). The Office cautions that a specification must do more than simply state that an improvement exists.
“Conversely, if the specification explicitly sets forth an improvement but only in a conclusory manner (i.e., a bare assertion of an improvement without the detail necessary to be apparent to a person of ordinary skill in the art), the examiner should not determine that the claim improves technology or a technical field.” (p. 2).
This addition suggests that boiler-plate paragraphs asserting “improved efficiency” or “better speed” will likely fail to support eligibility if unaccompanied by a technical explanation of how those results are achieved.
MPEP § 2106.05(a) – The “Reflects” Standard
Revisions to MPEP § 2106.05(a) reinforce the requirement that the claim “must include the components or steps of the invention that provide the improvement described in the specification” (p. 3). However, the memo offers a degree of flexibility: “The claim itself does not need to explicitly recite the improvement described in the specification (e.g., ‘thereby increasing the bandwidth of the channel’)” (p. 3).
Instead, the claim must be evaluated “as a whole” to determine if it “reflects” the improvement. The new text cites Desjardins to illustrate that improvements in “how the machine learning model itself would function in operation” are “not subsumed in the identified mathematical calculation” (p. 3).
Another added section cautions examiners that “When evaluating a claim as a whole, examiners should not dismiss additional elements as mere ‘generic computer components’ without considering whether such elements confer a technological improvement to a technical problem” (p. 4). This instruction explicitly links Desjardins to the evaluation of computer components or systems, preventing the improper discounting of algorithmic improvements.
MPEP § 2106.05(f) – Technological Solutions vs. Mere Instructions
The memo also revises the ninth paragraph of MPEP § 2106.05(f) to distinguish claims that provide a “technological solution to a technological problem” from those that merely instruct a computer to apply an abstract idea. Desjardins is now listed alongside other key Federal Circuit decisions that successfully integrated judicial exceptions into a practical application:
Internet Interactions: In DDR Holdings, claims were eligible because they specified how interactions were manipulated to yield a result that “overrode the routine and conventional sequence of events” (p. 4).
Content Filtering: In BASCOM, the court found a “technology based solution” of filtering content that overcame prior art disadvantages, rather than just an instruction to filter (p. 5).
Inertial Sensors: In Thales Visionix, the particular configuration of sensors and method of using raw data was deemed a “technological solution to a technological problem” that eliminated complications inherent in previous solutions (p. 5).
The USPTO now adds Desjardins to this line of cases, noting that the claims “reflected a specific improvement that addressed the technical problem of ‘catastrophic forgetting’ in continual learning systems, while allowing artificial intelligence systems to variously optimize system performance, use less storage capacity and reduce system complexity” (p. 5).
This categorization reinforces that AI claims addressing specific technical hurdles in model training should be treated as technological solutions, akin to the sensor configurations in Thales or the filtering architectures in BASCOM.
Examples of Claims Improving Technology
The memorandum provides a specific list of examples that the USPTO considers to be improvements in the functioning of a computer or other technology. These examples are now cited in MPEP § 2106.04(d)(1) and serve as safe harbors for eligibility arguments.
Examples of claims that improve technology or a technical field include:
Database Structures: “Data structure claims to a self-referential table for a computer database were directed to an improvement in computer capabilities and not directed to an abstract idea” (citing Enfish, LLC v. Microsoft Corp.) (p. 3).
Animation Rules: “Claims to automatic lip synchronization and facial expression animation were directed to an improvement in computer-related technology and not directed to an abstract idea” (citing McRO, Inc. v. Bandai Namco Games Am. Inc.) (p. 3).
Memory Systems: “Claims to an enhanced computer memory system were directed to an improvement in computer capabilities and not an abstract idea” (citing Visual Memory LLC v. NVIDIA Corp.) (p. 3).
Virus Scanning: “Claims to virus scanning were found to be an improvement in computer technology and not directed to an abstract idea” (citing Finjan Inc. v. Blue Coat Systems, Inc.) (p. 3).
Network Security: “Claims to detecting suspicious activity by using network monitors and analyzing network packets were found to be an improvement in computer network technology and not directed to an abstract idea” (citing SRI Int’l, Inc. v. Cisco Systems, Inc.) (p. 3).
Machine Learning Training: “Claims to a method of training a machine learning model were directed to improvements in the machine learning technology itself and additionally included data structure elements reciting adjustments in values to plurality of performance parameters while preserving prior values” (citing Ex Parte Desjardins) (p. 3).
Thoughts & Analysis
Benefits
The codification of Desjardins provides a powerful tool for AI applicants. By explicitly listing “reduced storage” and “reduced system complexity” (p. 2) as valid technological improvements for machine learning, the USPTO has countered the common examiner assertion that such efficiency gains are merely abstract or economic.
The addition of the Desjardins examples in xiii and xiv in MPEP § 2106.05(a) (p. 4) gives practitioners concrete citations to rebut § 101 rejections for model training inventions.
Challenges
To many practitioners, the bar for specification drafting feels like it has been raised. The explicit warning against “bare assertion[s]” (p. 2) requires inventors to disclose the specific mechanisms enabling the improvement.
Practitioners may need to gently prod inventors who try to keep the “secret sauce” of the training methodology hidden while still claiming the benefit. A vague specification will likely result in an eligibility rejection that cannot be cured by claim amendments alone.
Furthermore, stakeholders face the practical hurdle of inconsistent examiner implementation. Training a corps of thousands to distinguish between a “conclusory assertion” and sufficient “technical detail” is a significant undertaking.
There is a distinct possibility that Art Units will apply these standards unevenly, creating a landscape where eligibility depends as much on the assigned examiner as on the merit of the invention.
Many practitioners have seen “unpersuasive” too often to believe that the USPTO will be more limited with § 101 rejections any time soon. Until the Office establishes a consistent track record of allowance (or reasonableness) under Desjardins, applicants should anticipate a period of more unpredictability.
Risks
While Desjardins is binding on the examining corps, it does not bind the federal courts. There is a risk that a patent granted under this robust USPTO guidance could still be invalidated in litigation if a district court or the Federal Circuit takes a narrower view of what constitutes a “technological improvement.”
Additionally, the “reflects” standard remains subjective. An examiner might arguably agree that the specification discloses an improvement but disagree that the claim language adequately “reflects” it without explicitly reciting the result, despite the MPEP’s assurance that explicit recitation is not required. Maybe this is where the SMEDs come in.
Conclusion
The Desjardins memo represents a pragmatic step forward for AI patent eligibility at the USPTO. It confirms that improving the training process of a neural network is a technological improvement, not just a mathematical exercise. It codifies examples and explanation that could live long past an administrative changeover.
However, the accompanying MPEP revisions also appear to demand a higher level of specificity and precision in drafting. Practitioners who have not yet learned to dive deep in the technical waters when writing an application may need to take a swim lesson or two.
Patent owners should ensure that their applications do not merely claim the result of a software or AI process but technically describe and claim the specific (training) structures that achieve those results.
More USPTO training and communications on § 101 are certainly coming, and stakeholders will continue to keep their eyes and ears open.
Disclaimer: This is provided for informational purposes only and does not constitute legal or financial advice. To the extent there are any opinions in this article, they are the author’s alone and do not represent the beliefs of his firm or clients. The strategies expressed are purely speculation based on publicly available information. The information expressed is subject to change at any time and should be checked for completeness, accuracy and current applicability. For advice, consult a suitably licensed attorney and/or patent professional.