Beyond Application: The New ML Patent Paradigm
Authored by Babak Akhlaghi on May 20, 2025. The recent Recentive v. Fox ruling fundamentally changes how we think about machine learning patents. The Federal Circuit has drawn a clear line: applying generic machine learning techniques to new data environments without technological improvements is no longer patent eligible under 35 U.S.C. § 101.
This decision will have profound impacts on patent drafting and prosecution. We’ll likely see an increase in 101 rejections issued by examiners when reviewing AI-related inventions. To understand the implications, we need to examine what the court actually said. You can read my full summary of the case here.
Understanding the Court’s Position
We can’t interpret the Court’s assertion of “improving the machine learning itself” too narrowly. Such an interpretation wouldn’t align with the spirit of the Court’s holding. The Court specifically stated that “Machine learning is a burgeoning and increasingly important field and may lead to patent-eligible improvements in technology.” They also clarified that “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 key takeaway is that applying machine learning to an abstract idea doesn’t render the abstract idea non-abstract. Instead, the focus should be on how machine learning results in an improvement of a computer or a technology. You can demonstrate this in several ways.
First, if the improvement is to the actual model, you must show the specific improvement. Second, if the improvement is not to the model but to the input or output processing, you must describe those processes in detail. Broad functional recitation is insufficient. Claims need to describe how the improvement is achieved from a technical computer language standpoint.
Disproportionate Impact on Startups
This ruling will disproportionately affect smaller companies for several reasons. Startups have limited resources to fight against 101 rejections. Rather than traversing rejections or pursuing appeals, both costly options, they may simply abandon their patent applications.
It will also negatively impact fundraising efforts. Startups often use patents to maintain exclusivity, which may be their only asset when competing with resource-rich companies. However, their issued patents are now more likely to be subject to 101 attacks and invalidation by larger corporations.
Unless the 101 issue is resolved through much-needed legislation, the uncertainty surrounding it will likely result in companies pursuing open-source strategies for certain ML applications while only patenting those innovations that clearly result in improvements to computer technology.
Practical Strategies for Patent Drafting
Many AI-related claims qualify as improvements to computer operations or advancements in technological areas. Patent practitioners should ensure that a technical explanation of the improvement is clearly outlined in the specification and that the claims mirror this improvement.
When drafting AI-related inventions, the initial phase of the Alice/Mayo test typically examines whether the claims showcase specific advancements in computer capabilities or merely utilize AI tools for processing abstract ideas. Therefore, claims should emphasize technical aspects and describe functional mechanisms from a technical standpoint, rather than relying on broad functional descriptions.
Avoid generalizing statements that the invention works with all conventional ML models if it works best with a specific model. Instead, focus on specifics. In this case, the Court consistently referred to the specification which highlighted that the ML could be any conventional model.
When prosecuting such applications, engage with examiners early to understand what they consider patent eligible. The patent eligibility line at the USPTO gets redrawn daily, so early engagement can expedite prosecution and overcome 101 rejections more easily and at substantially reduced costs. Additionally, when responding to 101 rejections, build a strong written record during prosecution by referencing binding case law from the Federal Circuit and Supreme Court. This approach strengthens your case, providing a more robust foundation if the patent is later reexamined during litigation or an IPR.
Collaboration Between Technical and Legal Teams
Technical teams need to explain problems in technical detail to legal departments and protect the technical solution while identifying the advantages and how they’re achieved. Legal departments and outside counsel must push for more technical details about how the invention works.
If improvements are to the ML model itself, those improvements must be highlighted in the specification and figures and appropriately claimed. If the invention involves preprocessing inputs or postprocessing outputs, those features must be carefully described independent of the machine learning model and claimed alongside it.
The specification should outline with specificity how the claims result in a technical improvement to a computer or technology.
Looking Forward
Most inventions we’ll see will likely be directed to using existing ML models to address technical problems in specific fields. In many cases, the ML model itself isn’t being improved and, like in Recentive, could be any conventional model. However, the key innovation often lies in the preprocessing or postprocessing steps applied to the model’s inputs or outputs.
Make sure to highlight these steps with technical specificity and claim them along with the ML model. The definition of “abstract idea” remains unclear, and its application by the USPTO and courts is inconsistent. The only effective solution may be legislative intervention.
In the meantime, practitioners should carefully review new case law and directives from the Federal Circuit and USPTO, drafting and prosecuting applications accordingly. The specification should describe the technical problem and solution with sufficient specificity from a technical standpoint and explain how the solution achieves a technical objective.
Claims should track the technical solution outlined in the specification, ensuring that at least the technical effect is inferred. Remember, broad functional statements will not be sufficient for patent eligibility, whether AI-related or not.
Stay Informed with Expert Insights!
Want more valuable tips on navigating patent law and protecting your intellectual property? Join our free weekly newsletter for the latest updates, expert advice, and exclusive content straight to your inbox. Subscribe today and never miss out on essential information that can make all the difference in your patent journey! Subscribe to our Newsletter