Publications

Last month, in Recentive Analytics, Inc. v. Fox Corp., the Federal Circuit affirmed the Delaware district court’s dismissal of a patent infringement suit, holding that four machine learning-based patents were ineligible for protection under 35 U.S.C. § 101. This decision adds clarity into how courts evaluate subject matter eligibility of patents that involve artificial intelligence (AI) and machine learning. The key takeaway is that claims that merely apply machine learning to automate tasks or analyze data—even in novel settings—risk being classified as unpatentable abstract ideas unless they disclose specific improvements to the underlying technology or the machine learning itself.

Case Overview: Dismissal Under 35 U.S.C. § 101

Recentive sued Fox in Delaware district court alleging that Fox’s use of software for TV scheduling and content mapping infringed four of its patents. Recentive’s patents[1] claimed methods and systems of using machine learning (a form of AI) to dynamically generate event schedules and update network maps in real-time.

The Delaware district court granted Fox’s motion to dismiss, holding that the asserted patents were directed to ineligible subject matter under § 101. Recentive appealed, and the Federal Circuit has now affirmed.

Legal Framework: the Alice/Mayo Test

To assess patent eligibility, the Federal Circuit applied the two-step Alice/Mayo test established by the duo of landmark Supreme Court decisions, Alice Corporation v. CLS Bank International (2014) and Mayo Collab. Servs. v. Prometheus Lab’ys, Inc. (2012).

• Step one of the test asks whether the patent claims fall into one of the ineligible classes of inventions: laws of nature, nature phenomena, or abstract ideas. If so, the analysis proceeds to step two.
• Step two evaluates whether the claims contain an inventive concept—an element or combination of elements that is sufficient to transform the ineligible subject matter into a patent-eligible invention.[2]

Federal Circuit’s Decision and Analysis

Step One: Are the Claims Directed to an Abstract Idea?
Although Recentive argued that its application of machine learning is not generic because the claimed algorithms function dynamically and in real-time, the Court held that this amounted to nothing more than the abstract idea of applying generic machine learning techniques to solve a conventional problem. Even though the Court recognized that “Recentive’s methods unearth useful patterns that had previously been buried in the data, unrecognizable to humans,” it went on to note that “Recentive also admits that the patents do not claim a specific method for ‘improving the mathematical algorithm or making machine learning better’”. The Court further noted that neither the claims nor the specification described how such a technical improvement would be accomplished.

Therefore, at step-one the Court determined that the Recentive patent claims are directed to an abstract idea.

Step Two: Is there an Inventive Concept?
At step-two of the Alice/Mayo test, the Court flatly rejected Recentive’s assertion that its use of “machine learning to dynamically generate optimized maps and schedules based on real-time data” is an inventive concept sufficient to transform the abstract idea into a patent eligible application. The Federal Circuit agreed with the District Court which recognized “this is no more than claiming the abstract idea itself.”

In justifying its rationale, the Court distinguished Recentive’s patents, which merely claim the use of generic machine learning, from its previous holdings of patent eligibility in cases where the patents went beyond the abstract idea to claim specific and improved computer functionality.[3] In the present case, the Court explained that Recentive’s claims merely recite the application of generic machine learning and computing hardware without disclosing improvements to the machine learning models themselves or the computing technology. The Court explained that this use of generic machine learning failed to transform the abstract idea into patentable subject matter. Furthermore, the Court was unpersuaded by Recentive’s argument that the patents were eligible because they apply machine learning to a new field of use to more efficiently perform a task previously undertaken by humans.

Practical Tips—What Would Have Saved These Patent Claims

Fatal to Recentive’s case was a lack of disclosure of a specific technical improvement to the machine learning algorithm or an explanation of the specific way the machine learning technology was used to achieve such an improvement. General enhancements in speed or efficiency alone are insufficient to establish patent eligibility. To maximize the likelihood of surviving eligibility challenges, applicants may wish to consider the following practice tips.

1. Claim Specific Technical Improvements
   Deploying a known machine learning model, even in a new field, is insufficient unless the patent application describes a specific technical improvement in how the machine learning model itself works or leads to a technological advancement.
2. Consider the Entire Machine Learning Pipeline
   Applicants should consider innovations along the machine learning pipeline that may support a patent-eligible claim. For example, training data may be uniquely selected or pre-processed, or data storage might involve a specially structured database enabling faster or more efficient access. Additionally, the application may describe innovative processing at the input and output stages of the machine learning model. Therefore, focused claims directed to specific points of innovation may be more resilient to § 101 patent challenges.
3. Avoid Functional Language
   Applicants should replace functional results-oriented language with specific technical detail—such as describing how parameters are optimized, how new model architectures are used, or how integration with hardware enhances performance.

Conclusion

The Recentive case underscores the need for applicants to go beyond describing functional uses of machine learning and instead show specific technological innovation. Driving the machine learning engine is not enough. In short, simply powering ahead with general statements of functional improvements in performance will not get the applicant past the eligibility finish line. Instead, a patent application will need to describe—if not claim—the specific technical improvements. By clearly describing how machine learning is improved, whether through algorithmic enhancements, unique data processing techniques, or improvements to the training pipeline—applicants will be better positioned to navigate the road to a successful patent.

[1] The four patents-in-suit are US10,911,811, US10,958,957, US11,386,367, and US11,537,960.

[2] For the benefit of the reader we have oversimplified the framework of the Alice/Mayo test, as did both the District Court and the CAFC. However, it is recognized that the test comprises a more complex set of inquiries, particularly for inventions determined to claim abstract ideas at step-one. The step-two inquiry involves two further prongs, the explanation of which is beyond the scope of this article. The interested reader is directed to the USPTO Guidance on subject matter eligibility and the flow diagram outlining the Alice/Mayo test.

[3] See, e.g., Enfish, LLC v. Microsoft Corp., (Fed. Cir. 2016) (disclosing “a specific implementation of a solution to a problem in the software arts” for complex spreadsheet analysis); Koninklijke KPN N.V. v. Gemalto M2M GmbH, (Fed. Cir. 2019) (disclosing “a specific means or method that solves a problem in an existing technological process” for improved error checking in data transmission systems); and McRO, Inc. v. Bandai Namco Games America Inc., (Fed. Cir. 2016) (disclosing the use of specific computer techniques different from those humans use on their own to produce natural-seeming lip motion for speech).