AI and patents in the US: The time to safeguard your innovation is now

At the end of September 2025, just a few days after taking office, the new Director of the U.S. Patent and Trademark Office (USPTO), John A. Squires, issued a decision explicitly signaling his support for artificial intelligence (AI) innovations.

Over the past decade, the USPTO has taken a very restrictive stance toward anything that sounded like “software,” “algorithms,” or “mathematical models,” following Supreme Court rulings such as Alice v. CLS Bank. Many AI-related applications were rejected under 35 U.S.C. § 101 (Section 101 of Title 35 of the United States Code: lack of patent-eligible subject matter) on the grounds that they were mere abstractions.

In his decision Ex parte Desjardins, the USPTO Director overturned a previous rejection issued by the Patent Trial and Appeal Board (PTAB) and reaffirmed that the claims in question were patentable under 35 U.S.C. § 101. This marks an important recognition that AI inventions can indeed be patentable when they provide real technical improvements to system performance, rather than being automatically dismissed as abstract algorithms. The Director emphasized that categorically denying patent protection to AI would jeopardize the United States’ leadership in this strategic field. He also insisted that the true filters limiting a patent’s scope should remain the traditional requirements of novelty (35 U.S.C. § 102), non-obviousness (35 U.S.C. § 103), and adequate disclosure (35 U.S.C. § 112), rather than eligibility under 35 U.S.C. § 101.

This pronouncement reflects a shift in the USPTO’s approach: instead of rejecting AI applications outright for “lack of patentable subject matter,” the focus is now on how the invention technically improves a system or process. For applicants, this means that the drafting of the patent specification becomes crucial: the more clearly the technical advantages—such as efficiency, reduced complexity, or resource optimization—are described, the stronger the chances of success, especially when facing novelty and inventive step examinations.

In summary, the USPTO’s official stance now supports the following:

• Not using § 101 as a general barrier against AI.
• The focus of examination should return to the classic criteria: novelty (§ 102), non-obviousness (§ 103), and adequate description (§ 112).
• AI inventions are patentable when they can be shown to improve the functionality of a system, model, or technical process.

In other words, examination is shifting back to the more objective terrain of prior art, giving applicants greater room to argue, provided the patent specification clearly describes concrete technical advantages (reduced complexity, lower storage requirements, higher efficiency, etc.).

Is it now “easier” to obtain AI patents in the U.S.? Not exactly. The process is certainly less arbitrary than before, since applications are no longer blocked at the outset on the grounds of being “abstract algorithms.” However, the novelty and non-obviousness requirements still stand—and may even become more demanding. This makes it essential to draft highly technical applications that clearly articulate measurable advantages.

To make the most of the USPTO’s new approach, drafting AI patent applications now requires an exceptionally technical and strategic mindset. This is an area where H&A’s specialized team plays a key role in identifying, describing, and robustly protecting the genuine technical improvements that set an invention apart and maximize its chances of approval.

To that end, the key points we emphasize when working with innovators developing AI-based solutions are:

1. 1. Focus on real technical improvements
      It’s not enough to say “the model learns better.” You must describe what technical problem it solves and how: less memory usage, lower computational cost, greater efficiency in training or deployment, improved robustness against failures, and so on.

1. 1. Draft a detailed specification
      The application should include full technical descriptions of how the algorithms operate, how they differ from prior solutions, and concrete implementation examples.

1. 1. Highlight quantifiable advantages
      Including comparative data or examples is highly recommended; for instance, figures showing reduced memory use, faster training times, or lower inference latency. These numbers help demonstrate that the invention is not an “abstract concept,” but a measurable technical improvement.

1. 1. Be prepared for the §§ 102 and 103 battle
      Now that § 101 is no longer the main obstacle, examinations will focus more heavily on novelty and non-obviousness. It’s crucial to identify relevant prior art and clearly explain why the proposed solution would not have been obvious to a skilled person in the field.

1. Build the right approach from the start
   The description of the invention should present AI not as an end in itself, but as an enabling technology that improves a computer system, a communications network, an industrial process, a medical service, or any other technical application.

In short:

We must seize this new landscape, which offers greater chances of success than the pre–Director John Squires era.

At H&A, we believe that protecting these technologies today is the key to leading tomorrow. Once again, we stand beside those who are betting on artificial intelligence—helping transform their breakthroughs into strong, effective protection that ensures each innovation is shielded against both examination challenges and market competition.