Teaching Old Models New Tricks: The Persistence of Alice in Machine Learning Patent Claims

[1]

Under the Alice/Mayo framework, subject matter eligibility has long caused headaches for U.S. Patent and Trademark Office (“USPTO”) employees and practitioners alike, because the doctrine is difficult to apply and often inconsistently enforced.[2]  Although Congress has attempted to pass remedial legislation, none have become law.[3]  In the meantime, practitioners must rely on a patchwork of Federal Circuit decisions to navigate this convoluted area of patent law.[4]

In Recentive Analytics, Inc. v. Fox Corp.,[5] the Federal Circuit addressed “whether claims that do no more than apply established methods of machine learning (“ML”) to a new data environment are patent eligible.”[6]  The claims at issue related to using ML and generic computer parts to optimize television programming and to create network maps for live events.[7]  At Alice step 1, the court concluded that the claims were directed to an abstract idea because they recited the use of generic ML technology.[8]  Additionally, the court held that “iteratively train[ing] or dynamically adjusting” models did not constitute a technological improvement.[9]  The court reasoned that although the claims applied ML to a new environment, such an application does not improve the underlying technology.[10]  At Alice step 2, the court also found that the claims did not amount to significantly more than the abstract idea of generating event schedules and network maps through the application of ML.[11]  Therefore, the court held that claims applying ML to new environments without improving upon the ML models are not patent eligible.[12]

The holding in Recentive Analytics reveals important implications for patent practitioners drafting claims involving ML technologies.[13]  Practitioners drafting applications for ML technologies must ensure the claims capture a concrete technological improvement, not simply apply ML within a new domain or field of use.[14]  Because inventions using generic ML models in new technological environments are not eligible for patenting, other limitations of a claim must constitute the inventive concept.[15]

Practitioners, therefore, must make careful drafting decisions early in the patent prosecution process.[16]  First, they must make sure the invention is worthwhile for their client to patent by determining whether it improves upon a particular technology or is just an application of ML.[17]  Next, if the invention does improve a particular technology, the application should describe and claim the specific technological improvement, either in the ML model itself or in another technological component.[18]  If, however, the invention is an application of ML to a new environment, the practitioner should consider advising a client to protect the invention as a trade secret rather than pursuing patent protection.[19]

Recentive Analytics and the recent administrative decision, Ex Parte Desjardins,[20] highlight potential tension between the Federal Circuit and the USPTO.[21]  In Desjardins, Director Squires determined that a claim directed to “computing . . . an approximation of a posterior distribution over possible values of the plurality of parameters” was eligible.[22]  The opinion stated that the claimed invention was an improvement in “conventional [ML] systems by addressing challenges in continual learning and model efficiency by reducing storage requirements and preserving task performance across sequential training.”[23]  The opinion justifies this position by explaining that “categorically excluding AI innovations . . . jeopardizes America’s leadership in . . . emerging technology.”[24]  This decision suggests a policy shift in patent eligibility where the office will be more amenable to certain implementations of AI in patent claims.[25]

Even though the facts of Recentive Analytics would suggest that the claims at issue there would be ineligible under Desjardins, practitioners should take note of the evolving landscape of subject matter eligibility.[26]  It is possible that claimed subject matter found allowable by the USPTO under Desjardins guidance may nonetheless be invalidated by the Federal Circuit on similar grounds to the Recentive Analytics decision.[27]  For inventions with ML limitations, it is imperative to direct the claims to the improvements in ML within a new technological environment instead of reciting generic ML operating on generic computers in a new environment.[28]

As more inventions incorporate ML or artificial intelligence, there is a clear need for alignment in guidance from both the USPTO and the courts.[29]  While Recentive Analytics provides authoritative guidance on patents claiming ML modeling, administrative decisions such as Desjardins demonstrate evolving policy arguments for a more permissive view.[30]  Until the Alice framework is overruled by the Supreme Court or replaced by statute, practitioners will continue to rely on precedent set by decisions like Recentive Analytics for guidance on subject matter eligibility.[31]

References

[1] Photo by Ecliptic Graphic, A Computer Circuit Board With a Brain on It, Unsplash (Jan. 21, 2025), https://unsplash.com/photos/a-computer-circuit-board-with-a-brain-on-it-_jg8xh2SsXQ.

[2] See Jay P. Kesan & Runhua Wang, Eligible Subject Matter at the Patent Office: An Empirical Study of the Influence of Alice on Patent Examiners and Patent Applicants, 105 Minn. L. Rev. 527, 532–53 (2020) (outlining issues with subject matter eligibility after Alice).

[3] See Patent Eligibility Restoration Act, S. 1546, 119th Cong. (2025) (proposing legislation that attempts to remedy numerous patent eligibility issues).

[4] See Recentive Analytics, Inc v. Fox Corp., 134 F.4th 1205, 1216 (Fed. Cir. 2025) (holding that applying machine learning to new technological environments constitutes abstract idea).

[5] 134 F.4th 1205 (Fed. Cir. 2025).

[6] See id. at 1205, 1211 (describing issue presented).

[7] See id. at 1208–10 (providing factual background).

[8] See id. (describing Alice step 1 analysis).

[9] See id. (discussing whether application of machine learning constitutes improvement in technology).

[10] See id. at 1216 (reasoning applying machine learning to new environments is not improving underlying technology).

[11] See id. at 1215 (describing Alice step 2 analysis).

[12] See id. at 1216 (stating decision).

[13] See id. (holding machine learning application to new technological environments alone is not patent eligible).

[14] See Practical Steps for Patent Prosecutors Addressing 101 Rejections, Wolf Greenfield (Feb. 20, 2024), https://wolfgreenfield.com/articles/practical-steps-for-patent-prosecutors-addressing-101-rejections (outlining practical steps for drafting patent applications in light of USPTO subject matter eligibility guidance).

[15] See Recentive Analytics, 134 F.4th at 1216 (limiting holding to situations where “generic machine learning [is applied] to new data environments, without disclosing improvements”).

[16] See James J. DeCarlo & Samuel S. Stone, Federal Circuit: Machine Learning Patents

Ineligible in Recentive Analytics, Inc. v. Fox Corp., Greenberg Traurig (Apr. 21, 2025) https://www.gtlaw.com/en/insights/2025/4/federal-circuit-machine-learning-patents-ineligible-in-recentive-analytics-inc-v-fox-corp (providing practitioner guidance for claim drafting after Federal Circuit’s decision in Recentive Analytics).

[17] See Enfish, LLC v. Microsoft Corp., 822 F.3d 1327, 1336–37 (Fed. Cir. 2016) (stating that claiming improvements in existing technologies are not directed to abstract ideas); see also McRO, Inc. v. Bandai Namco Games America, Inc., 837 F.3d 1299, 1314 (Fed. Cir. 2016) (describing claim as improving existing technological process).

[18] See Enfish, 822 F.3d at 1336–37 (requiring improvements in technologies to overcome claims being directed to abstract ideas).

[19] See Intellectual Property Toolkit – Trade Secrets, U.S. Pat. & Trademark Off., https://www.uspto.gov/sites/default/files/documents/tradesecretsiptoolkit.pdf (last visited Mar. 21, 2026) (providing information and guidance on trade secret protection).

[20] 2024-000567 (P.T.A.B. Sep. 26, 2025).

[21] Compare Recentive Analytics, 134 F.4th at 1216 (determining application of generic machine learning to new environments does not constitute patent eligible subject matter) with Ex Parte Desjardins, Appeal No. 2024-000567, at 8–9 (determining improvements made to artificial intelligence models constitutes patent eligible subject matter).

[22] Ex parte Desjardins, No. 2024-000567, at 6–7 (describing claim language and agreeing that is patent eligible).  Specifically, Director Squires indicated the claim language of “adjust[ing] the first values of the plurality of parameters to optimize performance of the machine learning model on the second machine learning task while protecting performance of the machine learning model on the first machine learning task” constituted the improvement necessary to move the claim into subject matter eligibility.  See id. at 9.

[23] See id. at 7 (outlining argument for subject matter eligibility).

[24] See id. at 9 (describing policy-oriented approach for more inventions that use artificial intelligence).

[25] See id. (highlighting policy goals).

[26] See Recentive Analytics, 134 F.4th at 1208–11 (providing background on patents at issue and procedural history).

[27] See Ex parte Desjardins, No. 2024-000567, at 10 (proposing use of section 102, 103, and 112 as “appropriate tools to limit patent protection to its proper scope” rather than section 101).

[28] See Recentive Analytics, 134 F.4th at 1216 (setting forth precedential ruling).

[29] See Kathi Vidal, With Artificial Intelligence Speeding the Innovation Process, What Does that Mean for Invention and a Properly Balanced Patent System?, U.S. Pat. & Trademark Off. (Apr. 18, 2023) https://www.uspto.gov/blog/with-artificial-intelligence-speeding-the (stating that “AI now appears in 18% of all utility patent applications”).

[30] See Memorandum from Charles Kim, Deputy Comm’r for Pats., to Pat. Examining Corps., Advance Notice of Change to the MPEP in Light of Ex Parte Desjardins (Dec. 5, 2025), https://www.uspto.gov/sites/default/files/documents/memo-desjardins.pdf (notifying United States Patent and Trademark Office personnel about change in subject matter eligibility policy).

[31] See Recentive Analytics, Inc. v. Fox Corp., 134 F.4th 1205, 1216 (Fed. Cir. 2025), cert. denied No. 25-505, 2025 U.S. LEXIS 4546 (Dec. 8, 2025) (denying Recentive’s petition for writ of certiorari).