AI Patents Face Six-Fold Increase in Eligibility Rejections

Artificial intelligence innovators are facing a systemic hurdle at the U.S. Patent and Trademark Office, with AI-related applications far more likely to be rejected on eligibility grounds than non-AI filings. This disparity highlights a growing friction between legacy patent laws and the technical nature of machine learning.

According to an analysis published by Patently-O, the gap in rejection rates is substantial. Among patent applications that received their first office action in 2025, 42% of those containing AI drew a § 101 eligibility rejection. In contrast, only 6.9% of applications without AI faced the same rejection during their first office action.

The Eligibility Struggle

The USPTO has long struggled to apply the Alice/Mayo framework—the legal standard used to determine if an invention is a patentable process or merely an abstract idea—to the rapidly evolving field of AI. Because many AI inventions involve mathematical algorithms and data processing, they frequently trigger § 101 rejections, which claim the invention lacks a sufficient "inventive concept" to move beyond an abstract idea.

To combat these "conflicting and confusing" eligibility questions, USPTO Director John A. Squires has introduced Subject Matter Eligibility Declarations (SMEDs). These declarations are designed to allow applicants to provide explicit evidentiary support, helping examiners understand the specific technological advancements and practical applications of an AI invention rather than viewing it as a generic computational process.

Industry Implications

The high rate of § 101 rejections suggests a significant barrier for AI developers in the United States. When nearly half of all AI-related filings are flagged for eligibility issues at the first hurdle, it creates uncertainty for venture capital and corporate R&D budgets. This environment may chill investment in domestic AI hardware and software or force companies to shift their intellectual property strategies toward trade secrets rather than public patents.

The Path Forward

Industry observers are now watching to see if the implementation of SMEDs will meaningfully lower the rejection rate in subsequent office actions. While the tool provides a mechanism for better communication between inventors and the USPTO, it remains to be seen if it can overcome the fundamental tension between traditional patent eligibility laws and the inherent nature of machine learning models. For now, the data indicates that the path to a granted AI patent remains significantly more treacherous than for traditional technologies.