Machine Learning Patent Eligibility: What the Federal Circuit Held in Recentive v. Fox and Since

Machine learning patent eligibility is the question whether a claim using a trained model is directed to a “new and useful process, machine, manufacture, or composition of matter” under 35 U.S.C. § 101 or to an abstract idea, which the Supreme Court excludes from that section. The Federal Circuit took it up as “a question of first impression” in Recentive Analytics, Inc. v. Fox Corp., No. 2023-2437 (Fed. Cir. Apr. 18, 2025) (precedential), and held that claims that “do no more than apply established methods of machine learning to a new data environment” are not patent eligible (slip op. 10). More on AI and patents is on the Tech & AI hub.

Are AI inventions patent eligible?

The decisions covered here do not hold them ineligible as a class. Recentive concludes: “Today, we hold only 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 § 101.” The same paragraph observes that machine learning “may lead to patent-eligible improvements in technology” (slip op. 18).

How does the Alice test apply to machine learning claims?

As it applies to other software claims. In Alice Corp. v. CLS Bank International, 573 U.S. 208 (2014), the Supreme Court set out two steps: “First, we determine whether the claims at issue are directed to one of those patent-ineligible concepts” (at 217). If they are, the court looks for an “inventive concept” that makes the patent “significantly more” than a patent on the ineligible concept itself (at 217–18).

For software, the Federal Circuit asks at step one whether the claims focus on “the specific asserted improvement in computer capabilities” or on “a process that qualifies as an abstract idea for which computers are invoked merely as a tool.” Recentive applied that test to “software patents (which includes machine learning patents)” (slip op. 10, quoting Koninklijke KPN v. Gemalto, 942 F.3d 1143, 1149).

What did the Federal Circuit decide in Recentive v. Fox?

It affirmed the dismissal of Recentive’s infringement suit against Fox on four patents: U.S. Patent Nos. 10,911,811, 10,958,957, 11,386,367 and 11,537,960. Judge Dyk wrote for a panel with Judge Prost and Chief District Judge Mitchell S. Goldberg of the Eastern District of Pennsylvania, sitting by designation. The District of Delaware had granted Fox’s motion to dismiss under § 101 and denied leave to amend (slip op. 1–2, 9).

The “Machine Learning Training” patents (’367 and ’960) claim scheduling live events with a model trained on historical data; claim 1 of the ’367 patent requires “at least one of a neural network ML model and a support vector ML model” and iterative training. The “Network Map” patents (’811 and ’957) claim mapping live events to television stations using “a machine learning technique to optimize an overall television rating” (slip op. 3–6 and nn.2–3).

Recentive acknowledged below that “the patents do not claim the machine learning technique itself,” and that training “is required for any machine learning model” (slip op. 8).

Why were the Recentive claims abstract at Alice step one?

Because, in the panel’s words, “the only thing the claims disclose about the use of machine learning is that machine learning is used in a new environment”: event scheduling and network maps (slip op. 13). The specifications called for “any suitable machine learning technique,” listing random forests, regressions, neural networks and others, and the court cited that language to show that the technology was conventional (slip op. 11).

The court gave four further reasons. Iterative training and dynamic adjustment “are incident to the very nature of machine learning” and “do not represent a technological improvement” (slip op. 12). The claims “do not delineate steps through which the machine learning technology achieves an improvement” (slip op. 13). A new field does not help, because “[a]n abstract idea does not become nonabstract by limiting the invention to a particular field of use or technological environment” (slip op. 14, quoting Intellectual Ventures I v. Capital One Bank, 792 F.3d 1363, 1366). And doing a task once done by humans with “greater speed and efficiency” does not create eligibility (slip op. 15).

What did the court hold at Alice step two?

That nothing in the claims supplied an inventive concept. Recentive identified it as “using machine learning to dynamically generate optimized maps and schedules based on real-time data and update them based on changing conditions,” which the court called “no more than claiming the abstract idea itself” (slip op. 16).

The panel found no error in denying leave to amend, noting that Recentive “failed to propose any amendments or identify any factual issues that would alter the § 101 analysis” (slip op. 17). The Federal Circuit denied rehearing on 23 July 2025, and the Supreme Court denied Recentive’s petition for certiorari, No. 25-505, on 8 December 2025, according to the Supreme Court docket.

How do Enfish and McRO differ from Recentive?

Their software claims were held not directed to an abstract idea because they recited a specific way of improving a technology; neither case concerned machine learning. In Enfish, LLC v. Microsoft Corp., 822 F.3d 1327, No. 2015-1244 (Fed. Cir. May 12, 2016), Judge Hughes, with Judges Moore and Taranto, held database claims “directed to a specific improvement to the way computers operate, embodied in the self-referential table” (slip op. 12). The court did “not read Alice to broadly hold that all improvements in computer-related technology are inherently abstract” (slip op. 11).

In McRO, Inc. v. Bandai Namco Games America Inc., 837 F.3d 1299, No. 2015-1080 (Fed. Cir. Sept. 13, 2016), Judge Reyna, with Judges Taranto and Stoll, reversed a judgment on the pleadings that claims to automated lip synchronization of animated characters were ineligible. “It is the incorporation of the claimed rules, not the use of the computer,” the court wrote, that “improved [the] existing technological process,” and nothing in the patents’ background suggested animators had used rules of the claimed type (slip op. 24).

The panel found Recentive’s claims “materially different” from those in McRO; they disclosed neither “a specific implementation of a solution to a problem in the software arts” (quoting Enfish, 822 F.3d at 1339) nor, quoting Koninklijke, “a specific means or method that solves a problem in an existing technological process” (slip op. 13). It described McRO as upholding claims to “specific computer techniques different from those humans use on their own” (slip op. 15–16).

How have later Federal Circuit panels applied Recentive?

In two nonprecedential 2026 decisions, panels relied on Recentive to affirm ineligibility. In Rensselaer Polytechnic Institute v. Amazon.com, Inc., Nos. 2024-1725, -1739 (Fed. Cir. Feb. 24, 2026) (nonprecedential), Judge Dyk, with Judges Schall and Stark, affirmed summary judgment from the Northern District of New York against U.S. Patent No. 7,177,798. Its claims process natural-language queries using case-based reasoning, which the opinion calls “a type of machine learning or artificial intelligence” (slip op. 1–3).

The panel wrote that “[g]eneric use of AI without other parameters, such as ‘improving the mathematical algorithm or making machine learning better,’ is abstract” (slip op. 6). Rensselaer’s point that case-based reasoning was new to the field was one that “speak[s] to novelty, which is a separate inquiry” (slip op. 9). The panel said it “need not decide” whether adding non-conventional data to a conventional database is non-abstract, because the database’s other information types were shown to be conventional (slip op. 7).

In Dental Monitoring SAS v. Align Technology, Inc., No. 2024-2270 (Fed. Cir. July 7, 2026) (nonprecedential), Judge Lourie, with Judges Schall and Taranto, affirmed summary judgment from the Northern District of California that claims of U.S. Patent Nos. 11,049,248 and 10,755,409, which analyze dental-arch images with a “deep learning device,” are ineligible (slip op. 2). Two claims required training on “more than 1000 images of dental arches”; the panel wrote that training on a specific subset of data is “incident to the very nature of machine learning” (slip op. 4, 6, 10). It did not reach Align’s § 112 argument (slip op. 6 n.2).

What did the court leave open?

Claims that recite an improvement to a machine-learning model itself. Recentive’s holding covers claims that apply generic machine learning “without disclosing improvements to the machine learning models to be applied” (slip op. 18), and Recentive had “repeatedly conceded that it is not claiming machine learning itself” (slip op. 11). The opinion therefore had no claim to an improved model before it; its statement that machine learning “may lead to patent-eligible improvements in technology” is an observation, not a holding.

Both 2026 decisions are nonprecedential and turned on their records, including unrebutted expert testimony in Rensselaer (slip op. 8–9) and the patents’ own descriptions of the device in Dental Monitoring (slip op. 12–13).

What may practitioners take from these decisions?

The opinions suggest several considerations. Specification language permitting “any suitable machine learning technique,” or listing available networks, was cited as evidence that the model was generic (Recentive slip op. 11; Dental Monitoring slip op. 12). Statements at a hearing and in briefs about how machine learning works were quoted at step one (Recentive slip op. 8, 12).

Training-data limits, iterative training and real-time updating were not treated as technological improvements in Recentive or Dental Monitoring. On claim construction at the pleading stage, Recentive repeated that a patentee “must propose a specific claim construction or identify specific facts that need development” (slip op. 17–18 n.5, quoting Trinity Info Media v. Covalent, 72 F.4th 1355, 1360–61). Practitioners drafting or asserting machine-learning claims may consider how the claims and specification describe the way a model achieves an improvement, the point on which Recentive found both silent.

The other requirements for a patent are in the explainer on what makes an invention patentable; AI inventorship, a separate question, is in Can AI Be Named as an Inventor?; the court’s decisions are on the Federal Circuit hub.