The Patent Eligibility Landscape for AI and Machine Learning Inventions is Shifting
Written by Pooya Shoghi, Partner and AI Committee Lead at Lee & Hayes
Recent developments from both the Federal Circuit and the USPTO suggest a more favorable environment for applicants.
For over a decade, one of the most significant barriers to patenting AI and software inventions has been the threshold question of subject matter eligibility (SME). Since the Supreme Court’s 2014 decision in Alice Corp. v. CLS Bank, patent applicants in these fields have faced the challenge of demonstrating that their inventions amount to more than an abstract idea implemented on a generic computer.
Recent developments, however, suggest that the landscape may be evolving. Here are the key signals:
- The Federal Circuit acknowledged that AI model improvements may constitute patent-eligible technological advances. In Recentive Analytics v. Fox Corp., the court held that applying a generic machine learning model to a conventional task is not patent eligible. That holding was expected. What was notable, however, was the court’s repeated emphasis on what the claims lacked: namely, any specific method for improving the machine learning algorithm itself. That language, while not part of the court’s key holding, suggested that innovation in the model itself could constitute the type of technological improvement that renders a claim patent eligible.
- A precedential decision recognized improvements to a machine learning model as patent-eligible subject matter. In In re Desjardins, the USPTO’s Appeals Review Panel vacated an SME rejection of claims directed to training a machine learning model on multiple tasks while preserving the model’s ability to perform previously learned tasks. Director Squire found that these claims properly integrated an abstract idea into a practical application by reduced storage requirements, lowered system complexity, and the prevention of “catastrophic forgetting” (a classic problem in training AI models). The decision also cautioned against overbroad eligibility rejections, warning that categorically excluding AI innovations from patent protection jeopardizes American leadership in critical emerging technologies.
- The USPTO cautioned examiners against overbroad eligibility rejections in software and AI-related arts. In an August 2025 memorandum, the USPTO reminded examiners to apply the mental process grouping narrowly, to distinguish claims that recite a judicial exception from those that merely involve one, and to avoid oversimplifying claim limitations. The memo reinforced that eligibility rejections should only issue when ineligibility is established by a preponderance of the evidence, and not on the basis of uncertainty alone.
- The USPTO introduced a new evidentiary mechanism to support eligibility arguments. In a December 2025 memorandum, the Director designated Desjardins as precedential and issued guidance on Subject Matter Eligibility Declarations (SMEDs): a tool which allows applicants to submit expert evidence, including comparative testing data and performance benchmarks, to demonstrate that a claimed invention constitutes a genuine technological improvement. One of the USPTO’s own illustrative examples involves a novel neural network architecture with demonstrably superior performance over prior systems.
What this means for innovators: While the law remains unsettled, the trajectory favors applicants who can articulate with specificity how their AI innovations advance the underlying technology. Strategic claim drafting, capturing core machine learning improvements in independent claims while anchoring the invention to concrete technical applications in dependent claims, is more important than ever. Specifications should detail performance improvements with sufficient rigor to withstand eligibility challenges and to support a SMED if needed during prosecution.
If you are developing novel AI or machine learning technologies, now is an important time to evaluate your patent strategy. Our team regularly advises clients on navigating the evolving eligibility framework for AI inventions.
