In Recentive Analytics Inc. v. Fox Corp., No. 23‑2437 (Fed. Cir. Apr. 18 2025), the Federal Circuit issued its first precedential decision squarely addressing § 101 eligibility for machine‑learning (ML) patents. This opinion delivers a clear message to innovators and patent practitioners: merely applying a generic machine learning technique to a new field, without improving the underlying technology itself, is not enough to secure a patent.
Factual Background
Recentive Analytics, Inc. ("Recentive") sued Fox Corp. and its affiliates ("Fox") for infringing four patents. The patents were divided into two groups: the "Machine Learning Training" patents and the "Network Map" patents. Both groups claimed methods for using machine learning to optimize complex scheduling tasks in the media and entertainment industry.
- The Machine Learning Training patents, 11,386,367 (’367) & 11,537,960 (’960), claimed methods for dynamically generating and updating schedules for live events by training an ML model on various parameters like venue availability, ticket prices, and historical data.
- The Network Map patents, 10,911,811 (’811) & 10,958,957 (’957), claimed methods for creating and optimizing network maps, which determine which television programs are shown in specific markets at particular times to maximize ratings.
Recentive acknowledged that its patents did not claim an improvement to the machine learning algorithms themselves, but rather the "application of the machine learning technique to the specific context[s]". The district court held that the patents were directed to abstract ideas under § 101 and granted Fox's motion to dismiss. Recentive then appealed to the Federal Circuit.
The Court's Analysis: Generic ML on New Data is Abstract
The Federal Circuit affirmed the district court's decision, creating a significant precedent for AI/ML patent law albeit not unexpected based on current § 101 jurisprudence. The court's analysis, rooted in the two-step framework from Alice Corp. v. CLS Bank Int'l, provides a clear roadmap of what to avoid when seeking patent protection for AI/ML inventions.
Addressing the case as a "question of first impression," the court held that "claims that do no more than apply established methods of machine learning to a new data environment are patent ineligible". The court found that the patents merely used generic ML models as a tool to automate a process humans had long performed manually.
⚠ Caution: “Apply ML” Is Not Enough Applying a known tool (generic ML) to a new problem or data set, without improving the tool itself, is likely an abstract idea under § 101. The focus must be on a technical improvement.
The court’s reasoning can be broken down into several key takeaways:
| Issue | Court's Analysis | Practical Consequence for Patent Drafting |
| Generic ML Techniques | The patents recited using "any suitable machine learning technique" (e.g., neural networks, decision trees) on generic computers. This functional, result-oriented claiming was deemed abstract. | Avoid broad functional language. Instead, specify how the ML model is improved or configured in a non-conventional way to achieve the result. Detail the specific architecture or process that is inventive. |
| "New" Field of Use | Applying ML to a new field (event scheduling, network mapping) does not make an abstract idea patent-eligible. The court stated, "[a]n abstract idea does not become nonabstract by limiting the invention to a particular field of use or technological environment." | The "field of use" is not the invention. The specification must describe an improvement within that field that is rooted in technology, not just the application of technology to it. |
| Inherent ML Characteristics | Features like "iteratively training" the model or making "dynamic" updates based on real-time data were not inventive. The court noted these are incident to the very nature of machine learning. | Do not rely on standard, inherent functions of ML as the basis for an inventive concept. If the training or updating process is truly novel, the claims and specification must detail how it departs from conventional methods. |
| Speed and Efficiency Gains | The fact that an ML model performs a task faster or more efficiently than a human is not enough to confer patent eligibility. The court noted that speed and efficiency from using computers do not themselves create eligibility. | Focus on how the computer or the ML process itself is improved, not just on the improved outcome. For example, does the invention reduce processing load, use memory more efficiently, or solve a technical problem in a new way? |
Practice Tips for Drafting & Prosecuting AI/ML Patent Applications
The Recentive decision is a critical lesson for patent prosecutors. To draft robust AI/ML patents that can survive a § 101 challenge, practitioners should adhere to the following tips:
ℹ Key Prosecution Theme Focus claims and specification on technical improvements to model architecture, training pipeline, or inference efficiency — not on business goals the model helps achieve.
1. Focus on and Claim a Concrete Technical Improvement
The core of the invention cannot be the abstract idea of using AI/ML to solve a business problem. The specification must clearly articulate a technical solution to a technical problem. This could be:
- An improvement to the ML model/algorithm itself (e.g., a new neural network architecture, a novel activation function, a more efficient training methodology).
- A specific, non-conventional way of integrating the ML model into a larger system that improves the functioning of that system.
- A novel pre-processing or post-processing technique for the data that enables the ML model to operate in a technically improved way.
- Novel latency‑reduction techniques, edge‑device quantization (post-training quantization, quantization-aware training, etc.), enhanced interpretability mechanisms, etc.
- etc.
2. Avoid Functional Claiming; Detail the "How"
Claims that merely recite a function (e.g., "generating an optimized schedule") without specifying how that function is achieved in a novel, non-generic way are vulnerable. The claims and specification must work together to describe the specific implementation.
- Instead of: "using a machine learning model to optimize X."
- Consider: "a method comprising: [Step A of a specific data transformation], [Step B of applying the transformed data to a specifically configured neural network with layers X, Y, and Z], [Step C of generating an output based on a novel loss function]..."
3. Draft the Specification to Tell a Story of Technical Innovation
The background and summary of the invention sections should frame the problem and solution in technical terms. Explain the shortcomings of prior art computer systems or AI/ML techniques, not just manual or business processes. Describe how the invention provides a specific improvement over those prior art technologies. If need be, file a Rule 132 declaration during prosecution introducing factual evidence (experimental data, expert opinions, or secondary-consideration proof) to rebut a rejection or objection raised by the examiner (see 37 C.F.R. § 1.132).
ℹ The Inventive Concept is Key At Alice step two, the "inventive concept" must be more than the abstract idea itself. Relying on "using machine learning to dynamically generate optimized maps and schedules" was fatal for Recentive because that was the abstract idea. The inventive concept must be found in the specific, unconventional implementation details.
4. Tie the Invention to Improvements in Computer Functionality
If possible, link the invention to an improvement in the computer's operation. For example, does the invention allow the computer to process data more efficiently, reduce latency, consume less memory, or handle a type of data it previously could not? These are hallmarks of a patent-eligible technological improvement rather than an abstract process merely implemented on a computer.
5. Tie Claims to Specific Hardware or Data Flow, and Provide Empirical Evidence in the Specification
For example, disclose why a particular GPU/TPU topology or memory‑sharing scheme enables the model to function. Show how data pre‑processing pipelines overcome some constraint in the prior art. Include benchmark results, error‑rate reductions, or training‑time savings and contrast your approach with prior‑art ML models to demonstrate a technical leap.
- Keep detailed R&D records: lab notebooks and ablation studies can evidence an “inventive concept” at trial.
- Explain algorithmic novelty in prosecution to head off § 101 rejections early.
- Carefully review claim language: verbs like “receiving,” “processing,” and “outputting” are red flags when unaccompanied by algorithmic detail.
About the Author and Firm
This analysis is provided by Kama Thuo, PLLC, an engineering & technology law firm focused on patents, AI, and wireless telecom law. Roque Thuo is a patent attorney registered to practice before the U.S. Patent and Trademark Office and is licensed in Arizona and California. He is also a licensed Professional Engineer (Electrical) in Arizona, bringing a deep technical understanding to complex patent and AI/ML matters.
As AI continues to transform industries, securing strong patent protection is more critical than ever. Whether you are developing a novel algorithm or applying AI in a groundbreaking way, our firm has the technical and legal expertise to protect your innovations. Contact us to learn how we can help at www.kthlaw.com/patents.