Kirk A. Sigmon and Sydney Huppert
In April, the Federal Circuit offered important guidance on the subject matter eligibility of patent claims directed to artificial intelligence and machine learning (“AI/ML”) in Recentive Analytics, Inc. v. Fox Corp., No. 2023-2437 (Fed. Cir. April 18, 2025) (“Recentive Analytics”). While this case clarifies some issues, it largely aligns with the U.S. Patent and Trademark Office’s (“USPTO”) longstanding approach to AI/ML subject matter eligibility. It thus does not significantly alter our recommendations regarding AI/ML patents. That said, careful and strategic patent drafting remains essential.
The Recentive Analytics Decision
In Recentive Analytics, the Federal Circuit held that “patents that do no more than claim the application of generic machine learning to new data environments, without disclosing improvements to the machine learning models to be applied, are patent ineligible under [35 U.S.C.] § 101.”
The patents at issue in the case fell into two sets: the “Machine Learning Training” patents (directed to collecting data, training a machine learning model, receiving an output schedule from the trained machine learning model, detecting changes to the inputs, and iteratively generating new, further optimized output schedules) and the “Network Map” patents (directed to collecting data, creating a network map from that data, incorporating updates to the data by updating the map, and determining program broadcasts using the network map).
Applying Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208, 216 (2014),the Court noted that both sets of patents were “directed to ineligible, abstract subject matter,” characterizing them as using “generic machine learning technology” to generate “event schedules and network maps.” The Court did not see much value in the iterative training/generation steps of the “Machine Learning Training” steps, characterizing those steps as “incident to the very nature of machine learning.” The Court also noted that many of the steps performed by machine learning had been performed by “event planners” “manual[ly].” Though the Court’s analysis was brief, the Court noted that there was nothing “significantly more” to transform these abstract ideas into a practical application.
The Court did identify some examples of subject matter-eligible AI/ML patents, including those involving “improving the mathematical algorithm or making machine learning better” or, potentially, claims supported by a specification that details some form of a “technical improvement.” Along those lines, the Court noted that “a specific implementation of a solution to a problem in the software arts,” Enfish, LLC v. Microsoft Corp., 822 F.3d 1327, 1339 (Fed. Cir. 2016), and/or “a specific means or method that solves a problem in an existing technological process,” Koninklijke KPN N.V. v. Gemalto M2M GmbH, 942 F.3d 1143, 1150 (Fed. Cir. 2019), would likely suffice.
Strategic Recommendations Largely Unchanged
In our experience, USPTO examiners have already long since required more than an “application of generic machine learning to new data environments” to find subject matter eligibility. We’ve warned against filing “do this, but with a machine learning model”-type claims for many years, in no small part because USPTO examiners have explicitly informed us (e.g., during interviews) that they would reject claims of that nature as being directed to abstract ideas.
Given the Court’s ruling and our experience, many AI/ML-related patent applications are still likely to be considered subject matter-eligible. These generally fall into two categories.
First, patent applications directed to improving AI/ML itself are very likely to still be found subject matter-eligible. Such patent applications can be argued to improve machine learning processes (e.g., vis-à-vis the speed of model training, the accuracy of model output) and thus can easily be argued to improve the functioning of a computer. In our experience, these applications rarely face pushback under 35 U.S.C. § 101 because it can be easy to show how these applications improve the functioning of a computer.
Second, patent applications where AI/ML (generic or not) is but one component in a larger, subject matter-eligible system can still be found subject matter-eligible. After all, many patent claims recite allegedly “generic” computers and still can be found patent-eligible because, taken as a whole, they remain subject matter-eligible. That said, we do anticipate that applications of this type might face some additional pushback due to Recentive Analytics as USPTO examiners seek to determine where the line is drawn between claims that simply recite a generic implementation of AI/ML and claims that use AI/ML as part of some larger, subject matter-eligible system.
Key Takeaway: Defensive Drafting/Prosecution For AI/ML Patents Recommended
Given the Court’s ruling in Recentive Analytics, the value of careful patent application drafting cannot be overemphasized. The Recentive Analytics Court occasionally noted that the patents at issue in the case failed to provide sufficient detail to support the patent owner’s arguments that the patent applications described technical improvements. This could suggest a different result if, for example, the patents had been originally drafted defensively, with possible 35 U.S.C. § 101 rejections in mind. In turn, we recommend (and have long recommended) a defensive, § 101-minded approach to patent drafting and prosecution when handling AI/ML patents.
Posted: May 8, 2025
