USPTO Section 101 Reset: What Changed and Why It Matters for Insurance AI Patents

Three moves inside roughly six weeks reset patent eligibility for insurance AI. The USPTO rescinded the February 2024 AI inventorship guidance on November 28, 2025, revised the MPEP on December 5, 2025 to incorporate a precedential PTAB decision, and on December 8, 2025 the Supreme Court denied certiorari in Recentive Analytics, Inc. v. Fox Corp.

The result is a system pulling in two directions: better odds at the Office for carefully drafted new applications, and materially less litigation value in already-issued patents that claim machine learning broadly.

Key Takeaways

• December 8, 2025: the cert denial left standing the Federal Circuit's April 2025 holding that generic applications of machine learning to new data environments are ineligible under section 101. There is no corrective decision coming from above.
• Four patents were at issue in Recentive, and the claim structure the court rejected, collecting data, training a model, generating an output, updating in real time, describes a substantial share of insurance AI patents issued between 2018 and 2024.
• Four exposure channels run against an issued patent: continuation practice, reissue and reexamination, PTAB proceedings, and a section 101 motion at the pleading stage. They compound rather than substitute.
• Post-grant review reaches section 101 directly but only for nine months after issuance, so most insurance AI portfolios face the district court motion rather than the Board on eligibility.
• The MPEP revisions cut the other way. Sections 2106.04(d), 2106.05(a) and 2106.05(f) now push examiners toward eligibility where the claim as a whole reflects a technological improvement, even where the specification never uses that word.

What Changed at the Office and at the Court

The reset is bifurcated because the two halves came from different institutions, and only one of them binds a district court.

On November 28, 2025 the Office formally rescinded the February 2024 Inventorship Guidance for AI-Assisted Inventions, reframing AI systems as tools and declining to set a separate eligibility standard for AI-assisted inventions. A week later Deputy Commissioner Charles Kim issued an advance notice of MPEP changes incorporating Ex parte Desjardins, decided September 26, 2025 and designated precedential on November 4, 2025.

That is the pro-patent half. It sits alongside the August 2025 memorandum, which reminds examiners that the mental process grouping reaches only steps a person could practically perform in their head. Claims requiring a trained network, parameter updates at scale, or data volumes beyond human processing should not be characterized as mental processes. For insurance claims tied to policy-level exposure, claim-level development triangles or sub-segmented mortality tables, that is a real shield.

The restrictive half lives at the Federal Circuit and is now durable law. Under Alice step one the Recentive panel held the claims directed to abstract ideas implemented with known generic mathematical techniques; under step two it found no inventive concept, because the machine learning was functionally described and ran on generic equipment.

Why Enforceability Now Tracks Specificity, Not Breadth

The valuation consequence is the inversion: the broadest claims, which carried the highest notional value in 2020, are the ones a defendant can now attack cheapest.

Recentive, 134 F.4th 1205 (Fed. Cir. 2025), runs 19 pages and closes several escape routes at once. Iterative training on updated data does not create eligibility, being incident to the nature of machine learning. An abstract idea does not become concrete by limiting it to an environment, which forecloses the argument that applying models to underwriting or claims is itself inventive. Performing a human task faster does not confer eligibility, which undercuts specifications built around efficiency gains.

That reasoning sorts insurance AI claims along a specificity gradient rather than by subject matter:

• Highest exposure: collect insurance data, train a generic model, output a risk score, severity estimate or fraud flag, with no improvement to architecture, training method or data structure claimed. This maps onto the Recentive fact pattern directly. Defensive value survives; offensive value largely does not.
• Medium exposure: a specific technical mechanism is recited but described generally enough that a court can call it functional. Outcome turns on specification support and forum.
• Lower exposure: claims tied to non-generic computing improvements, such as reduced memory usage or training on volumes a person could not process, documented against a concrete metric. These look like Desjardins.
• Most defensible: claims integrating model outputs with a named actuarial calculation or regulated workflow, such as loss development triangles, credibility weighting, or reserve calculations under a stated methodology, with the specification also anchoring a technical improvement.

The cost asymmetry is what makes this a balance sheet question rather than a legal one. A section 101 challenge no longer waits for fact discovery. A motion to dismiss anchored in Recentive can end a case before the damages model is ever scheduled, which means an intangible carried at acquisition value may be worth defensive posture only. Most insurance AI applications draw an eligibility rejection during examination and issue anyway, so clearing the Office was never evidence a claim would hold up in court.

The Question Recentive Left Open

The framework above rests on a line the Federal Circuit drew but did not locate, and on a statute Congress has a live bill to replace.

The opinion does not resolve whether enhanced accuracy, efficiency or scalability can ever qualify as an eligibility-conferring improvement, or what magnitude of improvement would be required. Every tier boundary above depends on that answer. A portfolio sorted today is sorted against an unfinished rule.

The statutory floor is also provisional. The Patent Eligibility Restoration Act, introduced in the 119th Congress, would abrogate the Alice and Mayo framework and replace it with a statutory list of ineligible categories. If it advances, Recentive becomes a transitional artifact rather than the durable rule of the decade, and claims narrowed through reissue to survive it will have been narrowed against a standard that no longer applies.

Timing constrains the remedies too. Reissue and continuation are the vehicles for tightening claims to match what a specification actually teaches, but both run through examination under the current MPEP, so a family built on broad parent claims often cannot extend where it matters. Post-grant review, the one channel that reaches section 101 at the Board, closes nine months after issuance. For patents issued before late 2024 the eligibility fight happens in district court on the defendant's schedule, not the owner's.

Next in this series: Insurance AI Patent Enforceability After Recentive: Company-by-Company Analysis, applying the four-tier exposure framework above to the portfolios of AIG, Travelers, EXL, Quantiphi/Dociphi, and the 2018-2022 insurtech wave, with M&A diligence implications.