One Year Later, Thoughts on Recentive

It has been more than a year since Recentive Analytics, Inc. v. Fox Corp.,¹ was decided in which four Recentive patents covering machine-learning-generated TV broadcast schedules and network maps — claims directed to iteratively training a machine-learning model to identify relationships between event parameters and desired outcomes, generating schedules based on those parameters, and updating schedules in real time based on changing conditions — were found invalid under the §101 Alice patent-eligibility analysis. The Federal Circuit affirmed dismissal, framing this as a question of first impression: “whether claims that do no more than apply existing methods of machine learning to a new data environment are patent eligible.” The answer was no — patents that simply take an abstract idea and alter the mechanics in an AI environment are not patent-eligible subject matter.

Applying ML To New Field Not Enough

The Federal Circuit rejected Recentive’s argument that applying machine learning in a novel context and making tasks more efficient was sufficient, holding that “the application of existing technology to a novel database does not create patent eligibility.”

Iterative Training or Updating Inherent to ML, But Not Inventive

The court concluded that iterative training using updated data is incident to the very nature of machine learning and does not make machine learning itself better — the claims simply applied conventional ML techniques to a new data environment.

“Iteratively trained” or dynamically adjusted in the Machine Learning Training patents do not represent a technological improvement … . Iterative training using selected training material and dynamic adjustments based on real-time changes are incident to the very nature of machine learning.

Dyk, Circuit Judge, Recentive Analytics, Inc. v. Fox Corp., p. 13.

No Technical Improvement Disclosed

The court agreed the patents were merely directed to abstract ideas because they didn’t contain any actual technological improvement — Recentive admitted it neither invented nor improved any particular ML technique. The court specifically noted the claims “do not delineate steps through which the machine learning technology achieves an improvement,” and pointed out the MPEP counsels applicants to describe such an improvement in the specification.

Alice Step Two Resembles a §103 Inquiry

It is noted that the Alice step-two “significantly more” analysis in Recentive bears a striking resemblance to a nonobviousness analysis under §103 — the court essentially asked whether this was a novel use of AI, not just a novel application of it.

Recentive created an environment that is materially harsher for AI/ML patent holders. While the ruling did not hold AI/ML patents are invalid, it did make it harder to prosecute before the USPTO, as well as defend in federal court litigation or USPTO inter partes proceedings. Software patents that rely on generic computing functions or high-level algorithm references without concrete implementation details can expect the hardest treatment in either forum. As for practitioners, AI/software patents claiming mere automation of existing methods with generic computing are prime targets for early motions to dismiss on eligibility, and recently issued AI/software patents may be newly vulnerable to eligibility challenges. This decision underscores the challenges of securing patent protection for new applications of established machine learning techniques in various fields.

Combined, Recentive and GoTV Streaming v. Netflix,² both sustained § 101 invalidity findings against claims lacking specific grounding in technical improvement — the Federal Circuit has not held that AI inventions get any different treatment under the Alice/Mayo analysis than any other technology.

Under the current USPTO administration, examination of AI/ML is lessened in terms of the patent-eligibility scrutiny imposed on applicants during prosecution. Under Director Squires, the USPTO has recalibrated its examination approach to be more accommodating of AI-related inventions within the existing legal framework, including a memo (the Kim Memo, Aug. 4, 2025) reminding examiners handling AI/ML cases about eligibility standards. That memo instructs examiners that on a close call, they should only reject a claim if it’s more likely than not (>50%) ineligible, and should not reject merely out of uncertainty.

The current institutional climate at the USPTO supports more aggressive eligibility arguments, earlier allowance of AI claims with specific technical-improvement language, and strategic use of subject-matter-eligibility declarations to build a factual record — even as the Federal Circuit has not moved in the same direction. The USPTO’s guidance (Examples 47–49) is meant to help examiners apply eligibility standards specifically to AI inventions, and, per practitioners, is often more persuasive in day-to-day prosecution than Federal Circuit opinions since examiners are directly trained on it.

While patent examiners are currently leaning more permissive on initial allowance. Nevertheless, any AI/ML patent that does issue is on thin legal grounding based on the controlling case law (i.e., Recentive, GoTV), which collectively have held generic ML applied to a new domain will not survive an eligibility challenge in court. This is the path to invalidation.

Recentive is, at this point, final law. Recentive filed a petition for writ of certiorari before the Supreme Court just after the Federal Circuit denied the petition for rehearing en banc in July 2025, but that petition was denied in December 2025.

Practical Implications

AI/ML startups should expect a decent prosecution before the USPTO, but the expectation for third-party challenges or even federal court litigation is higher upon issuance of the patent. For practitioners, claims drafting must recite a concrete technical improvement, and not just restate a novel application of ML.

Additionally, AI/ML startups cannot rely on the old patent strategy of being the first to use ML for their particular area of industrial application. If it is. this alone will not survive a patent-eligibility challenge, however novel the business application.

As much as this confounds the analysis, because the Federal Circuit’s step-two eligibility analysis is functionally converging with a nonobviousness analysis, coupled with the USPTO’s own Appeals Review Panel (i.e., Ex Parte Desjardins) has directed patent examiners to lean on §§102/103/112 for potential rejections, rather than rely on §101 in these contexts, AI/ML startups should build a prosecution record showing genuine technical novelty. Director Squires’ Memo has re-confirmed this as of April 2026. This should not be novel use. Further, some ML innovations (e.g., model weights, training pipelines, data curation methods) may be better protected as trade secrets than as patents, since patenting requires public disclosure that may not survive litigated scrutiny.

For more information on §101 patent-eligibility, patent law in general, or any information relevant to your AI/ML startup, please contact Yonaxis I.P. Law Group.

---

Footnotes

1. 134 F.4th 1205 (Fed. Cir. 2025). ↩

2. See GoTV Streaming, LLC v. Netflix, Inc., ___F.4th___, Case No. 2024-1669, 2024-1744 (Fed. Cir. February 9, 2026). ↩