← Back to Software Patents After Alice

Ex parte Desjardins: When AI Training Claims Are Eligible

USPTO Appeals Review Panel, designated precedential 4 November 2025. An agency adjudication, not a court decision. Read from the decision; quotations verified against its text.

Holding

In a decision designated precedential on November 4, 2025 and written by the USPTO Director, the agency's Appeals Review Panel held that a claim to training a machine learning model integrates an abstract idea into a practical application where a specific claim limitation reflects an improvement to how the model itself operates, and warned that examiners "should not evaluate claims at such a high level of generality."

The Patent Office rejected a machine learning training application as an abstract idea. The Director convened a special panel, took the case back on its own motion, and reversed.

The reason is narrow and useful: the specification described learning new tasks without forgetting old ones, and a limitation in claim 1 actually required it, adjusting parameters to optimize performance on a second task "while protecting performance" on the first. That, the panel said, is an improvement to how the model operates rather than a mathematical calculation.

Two things get left out of most summaries. The Patent Office insists this announced no new practice. And the claims remain rejected for obviousness, so the applicant won eligibility and still does not have a patent.

Why an Agency Decision Sits Differently

This is a USPTO decision, not a court decision, and one Supreme Court case explains why that distinction now carries more weight than it used to.

On June 28, 2024, in Loper Bright Enterprises v. Raimondo, the Court overruled Chevron and held that the Administrative Procedure Act "requires courts to exercise their independent judgment in deciding whether an agency has acted within its statutory authority, and courts may not defer to an agency interpretation of the law simply because a statute is ambiguous."

There are no patent facts in that case. It matters here because the Patent Office's eligibility guidance, including the guidance this decision was applied under, is agency interpretation. Examiners still follow it, and it still governs what happens during prosecution. A court reviewing the same statute owes it less than it used to.

That is a statement about deference, not a prediction. Nothing in Loper Bright makes USPTO guidance invalid, unenforceable, or safe to disregard, and nothing in it makes AI patents easier or harder to obtain.

What This Case Is Not

Three things this is not. It is not a court decision; it is an agency adjudication. It is not a patent; the claims at issue stand rejected under § 103, so reporting the eligibility win alone leaves readers thinking one issued. And it is not a policy shift, a framing the USPTO's own memorandum contradicts.
  • It does not bind courts. Agency adjudications and Federal Circuit opinions are different authorities.
  • It does not mean training claims are eligible generally. The limitation in this claim did specific work.
  • It does not mean the applicant has a patent.

Educational, not legal advice. Whether any particular claim is eligible under § 101 is a legal conclusion a qualified patent attorney reaches on the actual claims.

Sources