A claim does not become patent eligible merely because machine learning performs the task. The claimed advance must be identified and expressed with technical precision.
In Recentive Analytics, Inc. v Fox Corp., decided on 18 April 2025, the United States Court of Appeals for the Federal Circuit considered four patents concerning machine learning used to generate television network maps and live event schedules. The court affirmed dismissal under 35 U.S.C. § 101.
What the court decided
The court described the claims as applying generic machine learning techniques to a particular environment. It found no claimed improvement to machine learning itself and no inventive concept that transformed the abstract idea into patent eligible subject matter.
The opinion treated the case as presenting a question of first impression. Claims that do no more than apply established machine learning methods to a new data environment are not patent eligible. This conclusion was tied to the claims and the specifications before the court.
Why the specifications mattered
The patents permitted the use of any suitable machine learning technology and listed familiar model types. The court also noted that iterative training and dynamic adjustment were ordinary incidents of machine learning on the record presented.
This is a drafting lesson. A specification that describes the model only at a functional level may make it difficult to establish that the claimed advance lies in computer technology rather than in the business result.
- What technical limitation existed in the prior system?
- Which data structure, training process or computing operation was changed?
- How does the change improve the operation of the computer or another technical process?
- What implementation detail separates the invention from routine use of a known model?
- Can the technical effect be measured and supported by examples?
Draft the mechanism, not only the desired result
Claims should identify the technical means that produce the improvement. Merely reciting collection of data, model training and generation of an optimized output may leave the claim focused on a result.
The description should support meaningful technical limitations. Relevant details may include model architecture, feature construction, memory use, signal processing, latency control, resource allocation or interaction with a physical system. The appropriate detail depends on the real contribution.
Do not read the decision too broadly
The court did not hold that machine learning inventions are categorically ineligible. Its analysis distinguished a specific improvement in computer capabilities from a process in which a computer is used as a tool. Eligibility remains claim specific and fact sensitive.
Practical implications for international applicants
An application drafted for several jurisdictions should disclose the technical problem, the technical mechanism and the demonstrated effect from the beginning. This provides a better basis for United States eligibility arguments and for European analysis of technical character and inventive step.
Frequently asked questions
Did Recentive make all AI patents ineligible?
No. The decision concerns claims that the court found used established machine learning in a new data environment without a claimed technical improvement.
Can detailed dependent claims help?
They can provide fallback positions if the details reflect the actual invention and are supported in the original disclosure.
Should applicants include experimental data?
Useful comparative evidence can strengthen the technical narrative. The required disclosure depends on the invention and the claims.