USPTO Updates MPEP § 101 Guidance for AI and Machine Learning Patents Following Precedential PTAB Decision

On December 5, 2025, the U.S. Patent and Trademark Office (USPTO) issued a memorandum from Deputy Commissioner for Patents Charles Kim to the Patent Examining Corps providing advance notice of revisions to the Manual of Patent Examining Procedure (MPEP), Ninth Edition, Revision 01.2024.

These changes impact how Patent Office examiners approach patent eligibility analysis under 35 U.S.C. § 101 for artificial intelligence (AI) and machine learning inventions. These updates incorporate guidance from the precedential PTAB Appeals Review Panel decision in Ex Parte Desjardins, Appeal No. 2024-000567 (PTAB Sept. 26, 2025; designated precedential Nov. 4, 2025), which vacated a § 101 subject matter eligibility rejection for claims directed to training a machine learning model.

According to the Desjardins decision, claims must be evaluated based on whether they are directed to a technological improvement rather than dismissed at a high level of abstraction as "merely reciting an abstract idea." The PTAB relied on Federal Circuit precedent, including Enfish, LLC v. Microsoft Corp. and McRO, Inc. v. Bandai Namco Games America Inc., emphasizing that software and in particular AI-based inventions can be patent-eligible when they improve how technology itself operates.

I. Key MPEP §101 Updates Affecting AI and Machine Learning Patents

The December 5 memorandum identifies several targeted revisions to Chapter 2100 of the MPEP that integrate Desjardins into USPTO examination practice:

MPEP § 2106.04(d), Subsection III (Step 2A, Prong Two). A new paragraph analyzes Desjardins, where the claims addressed the technical problem of “catastrophic forgetting”—the loss of previously learned information when a machine learning model is trained on new tasks. The PTAB found that the claimed training techniques enabled effective continual learning while reducing storage requirements and system complexity, integrating any abstract mathematical concepts into a practical application.

MPEP § 2106.04(d)(1). This section is revised to clarify that a specification need not explicitly label an invention as an “improvement,” so long as the improvement would be apparent to a person of ordinary skill in the art. Desjardins is cited as an example where improvements to machine learning training (such as parameter adjustments that preserve prior task performance) were credited, even though the claims did not expressly recite the improvement in those terms.

MPEP § 2106.05(a). The update emphasizes holistic claim evaluation under the broadest reasonable interpretation and cautions against dismissing claim elements as generic at too high a level of generality. Two new examples drawn from Desjardins are added: (xiii) claims directed to improved machine learning training that protects prior knowledge while learning new tasks, and (xiv) claims achieving performance enhancements through parameter adjustments in task-based models. As the PTAB explained, examiners and panels should not evaluate claims so generally that meaningful technical limitations are disregarded without adequate explanation.

MPEP § 2106.05(f). This section expands its discussion of “technological solutions” by incorporating Desjardins as an example of claims that solve a specific technological problem in AI—namely catastrophic forgetting—through concrete training techniques that improve performance and reduce system complexity.

II. Practical Impact for AI, Machine Learning, and Software Patent Applicants

These revisions are effective immediately and supersede prior MPEP guidance. Examiners are to apply § 101 patent eligibility analysis with focus on whether a claim integrates an idea into a practical, technological application, not analyses of novelty (§ 102), nonobviousness (§ 103), and enablement or written description (§ 112).

The updates also complement the USPTO’s December 4, 2025 memoranda addressing Subject Matter Eligibility Declarations (SMEDs) under 37 C.F.R. § 1.132, which encourage applicants to submit declaration evidence demonstrating technological improvements. Together, these developments are particularly significant for applicants in artificial intelligence, machine learning, software, and diagnostic technologies. The revised MPEP guidance provides clearer and more predictable pathways for overcoming § 101 subject matter eligibility rejections by emphasizing specification-supported technological improvements, practical applications, and concrete advances in computer and AI system functionality.

Have pending AI or machine learning related applications pending? Now is the time for an audit and re-calibration of prosecution. Have computer related, diagnostics, AI, or machine learning related inventions to protect? Contact us at CASTELLANO PLLC to discuss at admin@usaipr.com.

Related Resources

For additional context and strategic guidance, see:

• US Patent Alert: USPTO Issues Dual Memos on Subject Matter Eligibility Declarations - A Strategic Tool for AI and Diagnostic Innovators.

  
  

• US Patent Alert: USPTO Updates MPEP With Recent Subject Matter Eligibility Case Law

  
  

Frequently Asked Questions (FAQ)

How does the Desjardins decision affect AI patent eligibility under § 101?

The Desjardins decision confirms that AI and machine learning claims are more likely to be patent-eligible when they are directed to concrete technological improvements, such as improved model training techniques or enhanced computer functionality, rather than abstract ideas described at a high level.

Do AI patent specifications need to explicitly state that the invention is an improvement?

No. Under the revised MPEP guidance, an explicit statement of improvement is not required, provided that the improvement would be apparent to a person of ordinary skill in the art based on the specification and claims.

What is catastrophic forgetting, and why does it matter for patent eligibility?

Catastrophic forgetting refers to the loss of previously learned information when a machine learning model is trained on new tasks. In Desjardins, claims addressing this problem were found to solve a specific technological challenge in AI, supporting eligibility under § 101.

How should applicants respond to § 101 rejections after these MPEP updates?

Applicants should emphasize specification-supported technological improvements, practical applications, and concrete performance benefits. Where appropriate, evidentiary submissions such as Subject Matter Eligibility Declarations (SMEDs) may further strengthen the record.

Are these MPEP updates limited to AI inventions?

No. While particularly impactful for AI and machine learning, the revised guidance applies broadly to software-implemented inventions and other technologies evaluated under § 101.