In a closely watched decision, the Federal Circuit just made it clear: simply applying machine learning to a new domain—without improving the ML itself—won’t get you a patent.
In Recentive Analytics, Inc. v. Fox Corp. (No. 2023-2437, Apr. 18, 2025), the court affirmed the dismissal of four patents held by Recentive Analytics, all centered on using machine learning for TV broadcast scheduling and event planning. While the decision acknowledges the growing importance of AI, it draws a hard line on what qualifies as patent-eligible under 35 U.S.C. § 101.
What Was at Stake?
Recentive’s patents fell into two categories:
- Machine Learning Training Patents: Claimed systems for dynamically generating optimized live event schedules based on historical data.
- Network Map Patents: Targeted how television programs are assigned to channels across regions and times.
Their core idea? Replace manual, static scheduling with real-time, dynamic optimization using machine learning.
The Court’s Message: That’s Not Enough
The court emphasized that generic machine learning methods are “abstract ideas” unless they come with a specific technological improvement. Simply applying ML to a traditionally manual task, like TV scheduling, doesn’t cut it. In the court’s own words:
“Patents that do no more than claim the application of generic machine learning to new data environments… are patent ineligible under § 101.”
Why the Patents Failed
Here’s a breakdown of why the court ruled against Recentive:
- No New ML Tech
Recentive admitted it didn’t invent a new ML algorithm—it just applied existing methods to a new problem. The court saw this as no different than saying “do it with AI.”
- No Implementation Details
The patents didn’t explain how the ML models worked or improved the technology. They lacked any specific steps or mechanisms.
- Just a Field of Use
The court rejected the idea that applying ML to broadcasting makes it patent-worthy. Limiting an abstract idea to a specific industry doesn’t make it less abstract.
- Speed ≠ Innovation
Making a process faster or more efficient—especially with computers—doesn’t make it patentable. This principle has been upheld in multiple cases, and it applied here too.
- No Inventive Concept
Even at step two of the Alice framework (used to assess patent eligibility), the court found nothing “significantly more” than the abstract idea of using ML in scheduling.
Why This Matters for AI Patents
The Federal Circuit drew a clear line in the sand: if you’re using AI or machine learning, you need more than just an idea. You must show a specific, concrete technological improvement—not just automation of a manual process or use in a new domain.
The decision leaves open the possibility that patents could be granted for innovations that actually improve machine learning techniques. But just saying “we used AI” isn’t enough.
Final Thought
This case is a reminder that as AI becomes more integrated into various industries, patent law is holding firm on its standards. Innovators in machine learning must go beyond applying known techniques—they must push the technology itself forward if they want protection.