Patent Claims Running on AI? Federal Circuit Says Not So Fast on Patent Eligibility

The question of whether machine learning (ML)-based claims meet the subject matter eligibility requirements under current U.S. patent law remains hotly contested. The U.S. Court of Appeals for the Federal Circuit (CAFC) recently issued what appears to be the first precedential decision in addressing this in Recentive Analytics, Inc. v. Fox Corp. (Fed. Cir., 2025). The CAFC held that claims directed to merely employing established methods of machine learning to a new data environment are insufficient to render claims patent-eligible under 35 U.S.C. § 101.[i] The decision, which largely favored Fox, provides key guidance for patent owners and practitioners drafting and prosecuting ML-based claims, including consideration of one or more specific technological improvements in the underlying ML technology. In particular:

• Simply applying known ML algorithms in patent claims to a specific domain—even if performing a task previously undertaken by humans with greater speed and efficiency—may not satisfy patent eligibility.
• For applications relying on ML or other software-based technology to be patent-eligible, the patent claims must demonstrate a specific improvement to the underlying technology or present an inventive concept.

Recentive asserted four patents against Fox, alleging infringement of ML-based methods used to optimize event scheduling and network map generation for broadcast media. The asserted patents generally claimed processes that involve receiving various data inputs (e.g., venue availability, pricing and desired event metrics), training an ML model using historical data and using the ML model, based on user parameters and with a real-time updating function, to generate optimized schedules or network maps.[ii] Particularly, the ML model training step includes iteratively training the ML model to identify relationships within the data using historical event data.[iii]

The CAFC found the asserted claims ineligible under the Alice two-step framework.[iv] Specifically, for the first step, the CAFC concluded that the patents were directed to abstract ideas: the claims rely on the use of generic machine learning technology in carrying out the claimed methods for generating event schedules and network maps based on generic computing machines and processors.[v] The CAFC further concluded that the patents failed to provide an inventive concept that would transform the claims into patent-eligible subject matter.[vi]

The CAFC emphasized that the ML aspects recited in the claims—such as iterative training, applying defined inputs to a trained model and real-time algorithm updating for over time improvement based on the input —reflect generic machine learning functions and do not represent a technological improvement.[vii] The ML applications in this case are abstract in nature because the patents in question merely applied any suitable ML technique to specific fields of event scheduling and network map generation.[viii] Indeed, the CAFC held that the specification of the patents lacks a detailed description of how these ML techniques are implemented in the underlying claims.[ix] The CAFC went on to find that the claims at issue—which did not improve the ML technology itself but instead used it as a tool in a new context of broadcasting schedules —did not transform an abstract idea into patent-eligible subject matter, as the steps recited therein are incident to the very nature of machine learning.[x]

Based on the CAFC’s ruling, patent owners should consider drafting ML-based patent claims that do not reflect generic machine learning functions, and also include a detailed specification of the particular improvements to the underlying ML technology.

If you have questions about this case or if you would like to discuss strategies for patent prosecution, portfolio management or litigation, please contact the authors of this legal alert or your Dinsmore intellectual property attorney.