I’m pleasantly surprised by the decision in Ex parte Desjardins, Appeal 2024-000567, Application No. 16/319,040 (U.S. PTO Appeals Review Panel Sept. 26, 2025). I’ll be watching closely to see how this decision is integrated into the USPTO’s existing §101 guidance.

Although it’s often said that bad facts make bad law, independent claim 1 here looks more like a carefully constructed test case — one aimed squarely at probing patent subject matter eligibility in the context of training machine learning models. The Appeals Review Panel vacated the PTAB’s § 101 rejection, reasoning that the claims recited an improvement in how the model itself operates. The panel left intact, however, the PTAB’s §103 rejection.

The decision may signal a shift under Director Squires toward a more permissive posture on §101 in AI and machine learning contexts. That makes Desjardins worth watching not only for how the USPTO applies it, but also for whether the Federal Circuit eventually engages with similar fact patterns.

Finally, the case raises a practical enforcement question: even if patent claims on model training are found eligible, how do we actually detect infringement when the relevant conduct happens behind closed doors? The evidentiary and enforcement challenges surrounding AI patents may prove as consequential as the doctrinal ones.