An important U.S. court has ruled merely using generic machine learning does not constitute a new or useful process. This is a reminder that you need to be cunning in your patent strategy often looking out a decade to where the law will be.
Background
An important court for patent matters is the U.S. Court of Appeals for the Federal Circuit. Since 1982 the have been the court of appeals for patent matters. The decisions of the Federal Circuit for patent cases are binding precedent throughout the U.S. This is unlike the other courts of appeals whose purview is set by geography not subject matter leading to differing judicial standards depending on location, something known as circuit split. Decisions of a three judge panel at the Federal Circuit are only superseded by decisions of all judges of the Federal Circuit sitting as a group (“en banc”), a decision of the U.S. Supreme Court, or by applicable changes in the law.
Recentive Analytics, Inc., the owner of the patents, sued Fox Corp., for infringement of the patents. The district court dismissed, concluding that the patents were directed to ineligible subject matter under the Patent Act (35 U.S.C. § 101). Recentive appealed to the Federal Circuit.
The decision
Federal Circuit released its decision and reasons in a patent case involving AI. The panel affirmed a district court’s dismissal of the claims as ineligible, holding that “generic” machine learning technology is itself an abstract idea. See Recentive Analytics, Inc. v. Fox Corp., No. 2023-2437 (Fed. Cir. Apr. 18, 2025).
In its decision, the acknowledged the growing importance of machine learning, and exerted some care in limiting its holding:
Machine learning is a burgeoning and increasingly important field and may lead to patent-eligible improvements in technology. Today, we hold only that patents that do no more than claim the application of generic machine learning to new data environments, without disclosing improvements to the machine learning models to be applied, are patent ineligible under § 101.
The invalidity was ineligible subject matter
The three judges — Hon. Timothy Dyk, J.; Hon. Sharon Prost, J.; and Hon. Mitchell S. Goldberg, Chief District Judge, United States District Court for the Eastern District of Pennsylvania, sitting by designation — affirmed the decision of the lower court — District Court for the District of Delaware. They all hold the patents are invalid for not claiming patentable subject matter. Basically, the judges are saying the claims they looked at were not a “new and useful process, machine, manufacture, or composition of matter” (Section 101 of the Patent Act).
Decisions of this type follow a framework set out in Alice Corp. v. CLS Bank International, 573 U.S. 208 (2014) – the Alice framework. The Court concluded under step one that these patents claimed “ineligible, abstract subject matter” that is, producing event schedules and network maps using conventional and “generic machine learning technology”. For step two under the Alice framework, the Court nothing disclosed would transform the abstract ideas of generating event schedules and network maps using machine learning into something “significantly more” that would render the claimed subject matter patent eligible.
Our clients were prepared for this
We have been waiting for a decision like this. For about ten years now we have advised clients to patent more than simply their field plus machine learning. This advice was put into heavy rotation after 2022 when Chat GPT was released to the public. Some readers will remember conversations including “can’t we just” to which we replied “yes, and you’d better have a back-up position”.
We viewed the application of machine learning without going into details as being vulnerable to an attack under step one of the Alice framework. There is no satisfaction on being right here. The years since Alice have been hard on applicants and patentees.
What can you do to get a machine learning patent?
It is important to see the court’s reasons are limited in scope. The Federal Circuit did not hold that all inventions involving machine learning are patent ineligible. Indeed they suggested that improvements to machine learning methods and systems could be patent eligible.
You need to claim more than “the application of generic machine learning” to your data. Often it is a simple matter of considering how you apply a known technique or technology. We drill down into the details and then find the mix of details to claim with reasonable fall back position.
Your patent strategy needs to consider the law
Your patent strategy must consider the state of the law and the direction it is moving. You must think about what changes in the law will affect your pending and issued claims. This is especially so if you like to see terse claims. The brevity does imply breadth but often the claims are not resilient.
This is why you have an IP plan. Also this why you want to collaborate with your patent agent.
What does this mean if your company just uses artificial intelligence?
You may not be a developer of machine learning but may be a user. For example, a so called wrapper company applying machine learning to specific problems. You need to consider what to do carefully.
Be more than a generic machine learning applied to your domain. That is if your solution wraps or wraps itself known machine learning techniques, it likely will not qualify as patent eligible under Alice framework. The court explicitly stated that applying known machine-learning methods to new environments or datasets, without any technical improvement to the methods themselves, does not amount to patentable subject matter.
Details matter
Focus on the changes and improvements. You can make your invention patent eligible if you disclose and claim “specific technological improvement”. Often the small changes you make in using the machine-learning methods are technological. Any of the following should be included in your application new methods, specific model architectures, improved computational efficiency, or any change that fundamentally advances the underlying machine learning method not just its application.
Get into the details. Often it is a simple matter of considering how you apply a known technique or technology. We drill down into the details and then find the mix of details to claim. That is focus on the technical differences, and not the environment or the dataset. Clearly state in your patent applications how your technology advances or modifies existing machine learning technologies.
Places to look for details:
- New methods – Changes to the code that show improved results, efficiency (e.g., time, space, or energy), or functionality. And specifically look at how you handle your data.
- New structures – New ways of arranging or connecting the components.
- Hardware – Look for new uses, connections, and configurations of hardware.
Consider alternatives and additions
Consider alternatives to patents. You may have an innovation that isn’t helped by a patent application. Consider the alternatives like trade secret, speed to market as re-enforced by trademark rights, and the like.
Conclusion
We hope this post shows the importance of considering the law when drafting your patent strategy and patent claims. If you would like to protect your invention or improve your IP management practices. Please contact us at info@perpetualpatents.com.
