In April, the Federal Circuit issued a significant patent law ruling involving artificial intelligence. In , the Court addressed a core question facing many AI-driven businesses: When are solutions applying machine learning to real-world problems inventive and patentable? The Federal Circuit affirmed the trial court’s dismissal of the underlying case at the pleading stage under § 101 and held that applying generic machine learning models to scheduling and programming tasks—without disclosing any technological advances to the underlying machine learning techniques—failed to meet the eligibility standards under 35 U.S.C. § 101.
This past Wednesday, Recentive filed a combined petition for panel rehearing and rehearing en banc arguing that the April decision erased the line between § 101 and §§ 102/103 and erroneously imported § 112’s enablement requirement. Recentive’s petition also notes that several judges in the Federal Circuit have previously urged the full Court to review the boundary between eligibility and these validity requirements—suggesting that this case presented an excellent vehicle to do so.
Below, we break down the April ruling regarding the dismissal under § 101. We also discuss the implications for companies developing AI technology should the Federal Circuit deny Recentive’s rehearing requests.
The April 2025 Federal Circuit Decision
Recentive Analytics, Inc. asserted four patents related using machine learning to scheduling live events and assigning television programming. The patents fell into two groups: (1) the “Machine Learning Training” patents—focused on generating an optimized schedule for live events; and (2) the “Network Map” patents—focused on generating “network maps” that assign programming to TV markets to maximize viewership.
All claims involved collecting data, applying a machine-learning algorithm, and generating an output—either a schedule or programming map. The patents expressly disclosed that any standard machine learning techniques (e.g., regression, neural networks, decision trees, etc.) could be used to practice the invention. Importantly, Recentive acknowledged that the patents did not claim any machine learning techniques itself, but instead the application of known machine learning techniques to specific scenarios.
The Court found that the claims were simply directed to the abstract idea of using machine learning to optimize scheduling and programming and the claims were devoid of an inventive concept.
Recentive argued that the claims were actually directed to an improvement to technology, in part, because its claims required iteratively training the machine learning model. But the Court rejected that argument because it found that iterative training was embedded in the machine learning process.
Recentive also argued that its application of machine learning was not generic because it created a way for algorithms to function dynamically to allow the maps and schedules to be automatically customizable and updatable with real-time data. But the Court emphasized that neither the claims nor the specification disclose such an improvement.
Instead, the claims merely applied a generic computing technique—machine learning—to a particular use case—generating network maps and schedules. And the Court has long recognized that an abstract idea does not become nonabstract simply by limiting its field of use. Moreover, the Court explained—as a general matter—that patents merely using already鈥慳vailable technology as a tool to execute a claimed process, without more, were likely abstract.
Lastly, the Court found that Recentive’s articulated inventive concept—“using machine learning to dynamically generate optimized maps and schedules based on real-time data and update them based on changing conditions”—was essentially just claiming the abstract idea itself. And so, there was no tangible inventive concept to render the claims patent eligible.
Cautionary Signals for AI Patents
The April ruling tightens the standard for patenting AI-related inventions, particularly when they rely on off-the-shelf ML models. Here is how the April ruling would affect practitioners and innovators, should the April ruling remain upheld:
“Novel Method Using AI” Isn’t Enough – Merely claiming that a process uses machine learning—even iteratively or dynamically—won’t confer patent eligibility if the underlying method is conventional and the ML application is generic.
Claims Must Show Technical Improvement – To pass § 101, AI patent claims should demonstrate a specific technological advancement. Practitioners should try to understand the problems that needed to be solved—beyond just the business goals. Rather than simply asking how the AI fixed a business problem, practitioners should focus on the technological hurdles in implementing and incorporating the AI, which solve technological problems.
Reading between the lines, that means under the April ruling these sorts of claims are more likely to pass muster:
- A novel ML training method
- A novel ML data architecture or processing method
- Improved performance of the computing system executing the ML algorithm
Functional Results Won’t Save Abstract Claims – The April ruling shows that the panel remains skeptical of claims that only recite high-level results (e.g., “maximize rating,” “optimize schedules”) without describing how those outcomes are achieved in a non-abstract, inventive way. But regardless of whether the rehearing is granted, whether the claims recite the how should be a lingering question in every practitioner’s mind when it comes to AI patents.
In its recent rehearing petition, Recentive argues that under the April decision, no application of pre-existing machine learning models would be patent eligible and as a result would stifle investment in world-changing machine learning inventions. Whether the Federal Circuit will grant the petition for rehearing remains to be seen. But given the stakes and the issues at play, this is certainly a case worth watching closely. Stay tuned!