USPTO Encourages AI Innovation as Federal Circuit Exercises Caution

With the continued rise of generative artificial intelligence (AI), the question of how the United States Patent and Trademark Office (the “USPTO”) and courts will apply patent-eligibility requirements to AI has been at the forefront. The USPTO and the judiciary have begun to provide insights from which an answer may be gleaned. While the USPTO and the Federal Circuit both apply the same legal framework, they appear to have different views on the barrier that patent-eligibility requirements should present to inventions involving AI.

The USPTO, under new leadership, has strongly encouraged AI innovation and investment to signal that the United States will be a world leader in this novel field of technology. On the other hand, the Federal Circuit seems to be taking a more cautious approach that does not indicate preferential treatment for AI and machine learning. The difference lies not in the law itself but in how each institution signals to patentees, challengers, and the world at large about the future of AI in the United States.

• In Recentive Analytics v. Fox, the Federal Circuit held that applying generic machine-learning techniques to a new field does not make a claim patent eligible under 35 U.S.C. § 101, since such claims are simply directed to an abstract idea that contained no inventive concept.¹ Recentive reframes rejection under Alice step two, “merely requir[ing] generic computer implementation,”² by replacing “computer” with “machine learning.” The court did not treat iterative training and dynamic adjustments as technological improvements to machine learning since they are “incident to the very nature of machine learning.”³
• John A. Squires, in his first week as Director of the USPTO, led an Appeals Review Panel rehearing that vacated a PTAB-initiated § 101 rejection of a DeepMind continual-learning claim in Desjardins, holding the claim teaches improvements in training the machine learning model itself and should be assessed primarily under 35 U.S.C. §§ 102, 103, and 112 instead of § 101.⁴ The decision concluded that the claims were patent-eligible because they integrated a mathematical concept into a practical application—they changed how the model learns across tasks. While the claims pass muster under § 101, the panel allowed the requirements under § 103 to be the gatekeeper as to whether the AI claim was patentable. In this instance, it agreed with the PTAB that it was not.

Interestingly, both decisions highlight how the analysis under Alice step two may blur into obviousness under § 103. In Recentive, the Federal Circuit’s analysis as to whether the machine learning limitations did “significantly more” than the abstract idea bore a striking resemblance to the analysis for nonobviousness under § 103. The court, in essence, inquired whether this was a novel use of AI. The Appeals Review Panel in Desjardins instructed examiners to use §§ 102, 103, and 112 instead of § 101 in such contexts and maintained the rejection under § 103.

Despite the different outcomes, both the Federal Circuit and the USPTO approach patent eligibility for machine learning claims under 35 U.S.C. § 101 in much the same way. The Alice framework still stands, and the decisions in Recentive and Desjardins hinge on how patent eligibility for machine learning requires a claim to expressly recite a specific way that improves how the model learns or operates. As Recentive has shown, the same considerations for drafting claims implementing an “ordinary computer” should now be applied to “ordinary machine learning.” Whether a machine-learning claim is patent-eligible subject matter will rely on the language showing practical integration of an abstract idea to show technical improvements to the model itself.

The development to watch will be whether the judiciary leans toward a view that aligns with that clearly enunciated by the new leadership of the USPTO. In Desjardins, the Appeals Review Panel, led by Director Squires, goes out of its way in dicta to admonish examiners for “categorically excluding AI innovations from patent protection in the United States” since that “jeopardizes America’s leadership in this critical emerging technology.”⁵ Meanwhile, the Federal Circuit’s decision in Recentive reflects continuity with previous policies and approaches that appears inconsistent with the views of the current USPTO administration.

Ultimately, these two decisions indicate that the United States is on the cusp of more policy shifts. As Director Squires said in his opening remarks: “[f]rom crypto and AI to quantum computing and diagnostics, the marketplace is filled with breathtaking opportunities for invention and investment. Who knows what tomorrow will bring?” Considering the roles of the USPTO and the Federal Circuit in developing the law of the land, between the two, the latter will have the final say in the policy that the US implements. Thus, Director Squires may need to seek help from Congress to achieve his desired outcome should Federal Circuit decisions continue to place patent eligibility as a barrier to patent claims directed to AI.

¹ Recentive Analytics, Inc. v. Fox Corp., 134 F.4th 1205, 1208 (Fed. Cir. 2025).

² Alice Corp. Pty. v. CLS Bank Int’l, 573 U.S. 208, 134 S. Ct. 2347, 2350, 189 L. Ed. 2d 296 (2014).

³ Recentive, 134 F.4th at 1212.

⁴ Ex parte Guillaume Desjardins, et al., No. 2024-000567 (P.T.A.B. Sept. 26, 2025).

⁵ Id. at 9.