Federal Circuit's Section 101 Crackdown Hits Insurer AI Patent Moats

State Farm, USAA, and Allstate hold 77% of all insurer AI patents filed since 2014, a 680-patent portfolio built largely on the claim pattern the Federal Circuit invalidated three more times between February 6 and February 24, 2026 (Evident, Dec. 2025; Fed. Cir., Feb. 2026). Each ruling extended the same standard the court set in Recentive Analytics, Inc. v. Fox Corp. in April 2025: applying generic machine learning to a known problem, without a concrete technical improvement, is not patent eligible.

77%

Insurer AI Patents Held by 3 Carriers

Fed. Cir. Section 101 Invalidations Since Apr. 2025

Claims Invalidated in One Feb. 2026 Ruling

31%

Insurer Filings That Are GenAI, Not Traditional ML

Three Rulings in Eighteen Days

Between February 6 and February 24, 2026, the Federal Circuit handed down three separate Section 101 opinions, and all three reinforced the same rule: a patent that takes a known business or engineering problem and solves it with machine learning, without describing a specific improvement to the model or the pipeline, is an abstract idea ineligible under 35 U.S.C. § 101.

On February 6, the court affirmed summary judgment in Innovaport LLC v. Target Corp., No. 2024-1545, invalidating all 55 asserted claims across six related patents covering in-store product location systems. Three days later, in the precedential GoTV Streaming, LLC v. Netflix, Inc., No. 2024-1669, Judge Taranto reversed a jury infringement verdict and invalidated three patents covering methods for tailoring television content display to a wireless device's capabilities. The panel found the claims directed to the abstract idea of using a generic template tailored to user-specific constraints, and rejected GoTV's argument that collecting and combining data for a rendering device amounted to a technical advance.

The third ruling is the one insurers should read most closely. On February 24, in Rensselaer Polytechnic Institute v. Amazon.com, Inc., No. 2024-1725, a panel led by Judge Dyk affirmed invalidation of U.S. Patent No. 7,177,798, a natural-language patent asserted against Amazon's Alexa. The court's language cuts directly at the AI claim-drafting pattern common across insurance patents: "the generic use of AI without other parameters, such as 'improving the mathematical algorithm or making machine learning better,' is abstract" (Fed. Cir., RPI v. Amazon, Feb. 2026). Dennis Crouch, summarizing the trio for Patently-O, noted that patents drafted before the Supreme Court's 2014 Alice Corp. v. CLS Bank decision "were written in a different era of patent law" and are proving especially vulnerable now (Patently-O, Feb. 2026).

Each of the three cases had priority dates well before the current AI patent wave, ranging from 1999 to 2007. But the claim architecture they share, collect data, apply a known computational technique, output a result, is structurally identical to the "collect insurance data, train a model, output a risk score" pattern that dominates the insurer AI patent filings of 2018 through 2022. That is the connection the Federal Circuit itself drew back to Recentive, and it is the connection that matters for carrier IP strategy in 2026.

The State Farm-USAA-Allstate Concentration and What It Actually Claims

Evident's Insurance AI Patent Tracker, released in December 2025, quantified a pattern the industry had suspected for years: three P&C carriers, State Farm (326 patents), USAA (218), and Allstate (136), account for 77% of every AI patent filed by an insurer since 2014, a combined 680 out of roughly 883 tracked filings (Evident, via Insurance Journal, Dec. 22, 2025). We covered the concentration itself, and what each carrier is patenting, in a companion piece earlier this year. What that analysis did not test is how those claims read against the judicial standard now being applied to them.

A sample of issued patents from the largest filer illustrates the exposure directly. State Farm's U.S. Patent No. 11,861,470, titled "Simplistic machine learning model generation tool for predictive data analytics," and U.S. Patent No. 11,769,213, "Method of controlling for undesired factors in machine learning models," both describe methods for building and constraining ML models for insurance predictive analytics in functional, model-agnostic terms rather than tying the claims to a specific architecture or training modification. That is precisely the drafting posture the Federal Circuit has now rejected four times over: Recentive rejected iterative training on updated data as an eligibility-conferring feature, and RPI v. Amazon rejected "generic use of AI without other parameters" outright. Claims-triage patents, which route incoming claims by complexity and urgency, and risk-pricing patents, which output a severity or fraud score from policy data, sit in the same structural bucket. These are the two categories Evident identifies as the largest concentrations in the insurer AI patent base, and they are the categories built on the claim architecture the courts keep invalidating.

None of this means the underlying 680-patent portfolio is worthless. Defensive value, deterring casual copying, signaling R&D scale to investors and regulators, survives regardless of Section 101 exposure. What has changed is offensive value: the ability to sue a competitor or an AI vendor and expect the claim to survive a motion to dismiss.

The Insurance AI Patent That Survived

Specificity, not subject matter, is what separates a claim that survives from one that does not. The clearest insurance-specific test of that principle predates the February rulings but points the same direction. In Aon Re, Inc. v. Zesty.ai, Inc., No. CV 25-201 (D. Del. July 15, 2025), a federal judge denied ZestyAI's motion to dismiss Aon's infringement claims covering property-risk assessment from aerial imagery. The court acknowledged that Recentive bars "the mere application of generic machine learning to new data environments," but found the asserted patents recited "the patent-eligible arrangement of two independently trained classifiers to analyze property characteristics and conditions" (D. Del., Aon Re v. Zesty.ai, July 2025). Claim 1 of the patent at issue requires a specific classifier sequence that analyzes pixels from aerial images in a defined order to generate a risk score, a concrete technical arrangement rather than a functional description of an outcome.

The contrast is instructive for portfolio triage. Aon's surviving claim and State Farm's 11,861,470 both describe systems that produce an insurance risk output from a trained model. The difference is that Aon's claim specifies how the model is structured (two classifiers, applied in sequence, to defined image data) while the generic claim specifies only what the model does. Post-Recentive, and now post-RPI-v.-Amazon, that structural difference is the entire ballgame.

Claim Pattern	Example	2026 Judicial Exposure
Generic ML applied to a known insurance task (claims triage, severity scoring, pricing)	State Farm US 11,861,470; US 11,769,213	High. Matches the pattern invalidated in Recentive, GoTV, Innovaport, and RPI v. Amazon
Specific multi-model or classifier arrangement tied to defined data	Aon Re's aerial-imagery classifier patent (survived motion to dismiss, D. Del. 2025)	Lower. Survived direct application of Recentive
Agentic multi-agent coordination architectures	USAA's agentic underwriting and claims patents	Untested. No Federal Circuit ruling yet addresses multi-agent claim structures directly
Generative AI content generation and enhancement (aerial imagery clarification, document drafting)	USAA's GenAI aerial-imagery enhancement patents	Moderate. Depends on whether the claim specifies a concrete generation mechanism or a functional outcome

The taxonomy shift in the underlying filing data reinforces the split. Generative AI patents rose from roughly 4% of insurer AI filings in the early 2020s to 31% by October 2025 (Evident, Dec. 2025), a category Evident's own classification treats as functionally distinct from the traditional machine-learning claims that dominate risk modeling and pricing. Risk modeling and pricing patents remain, in Evident's framing, anchored to traditional machine-learning approaches, the exact profile the Federal Circuit has now voided on four separate occasions. Newer generative AI and agentic filings carry more implementation-specific claim language by necessity, since the underlying architectures (retrieval pipelines, multi-agent orchestration, classifier chains) are newer and less standardized, which gives drafters more to point to as a concrete technical mechanism. That does not make the newer filings automatically eligible; it makes them more likely to look like Aon's surviving claim than State Farm's exposed one.

What Weaker Patent Moats Do to the Vendor Build-vs-Buy Calculus

Patent portfolios function as competitive moats by raising the cost of replication. A carrier that patents a specific claims-triage method can, in theory, block a rival carrier or an AI vendor from shipping a similar system without a license. That threat is what keeps some carriers building internally rather than adopting a shared vendor platform, and it is exactly the threat that generic, functionally drafted claims can no longer reliably deliver.

The vendor side of the market is not waiting to find out. Sixfold raised a $30 million Series B in January 2026 with Guidewire as a strategic backer, and Munich Re has already integrated Sixfold's underwriting AI directly into its Realytix Zero platform, spanning submission intake through pricing and binding (fintech.global, Jan. 2026). Applied Systems acquired Cytora in September 2025 specifically to fold Cytora's AI-enabled risk-processing platform into its own broker and carrier product suite (InsurTech Analyst, Sept. 2025). Neither deal was priced as if the acquirer feared a patent infringement suit from State Farm, USAA, or Allstate over shared claims-triage or risk-scoring functionality. If the underlying carrier claims cannot survive a Section 101 motion to dismiss, that pricing behavior is rational rather than reckless: the licensing risk that would otherwise justify building a proprietary, patent-clear system in-house has fallen, and buying a vendor platform that performs a similar function carries less latent liability than it did in 2022.

The calculus is not uniform across claim categories. A carrier evaluating a vendor's generic claims-triage or pricing automation tool can reasonably discount the risk that a rival's Tier 1 patent (generic ML applied to a known problem) blocks the purchase. A carrier evaluating a vendor's agentic multi-agent underwriting system, the category USAA leads and Evident expects to see accelerate through 2026, faces a less settled picture, because no Federal Circuit panel has yet ruled directly on multi-agent coordination claims. Build-versus-buy decisions in that category should weight patent risk more heavily until the courts catch up to the architecture.

The M&A Diligence Discount

Patent portfolios are not just competitive tools; they are balance sheet items. Intangible assets, including patents, now account for roughly 90% of S&P 500 market capitalization and are estimated at more than $62 trillion globally, and undetected invalidity risk in a target's patent book is a recognized deal-killer in technology and life-sciences M&A diligence (DrugPatentWatch, 2025). Insurance carrier and insurtech M&A has not historically treated AI patent portfolios as a major line item in that analysis, in part because the portfolios themselves are recent. That is changing as carrier AI patent books grow large enough to show up explicitly in target valuations and as reps-and-warranties insurance underwriters price IP representations more carefully.

The practical effect of four straight Section 101 invalidations touching the exact claim pattern insurers favor is a compression of the multiple diligence teams should assign to an "AI patent moat" line item built on Tier 1, generic-claim patents. A due diligence team pricing a target's claims-triage or pricing-automation patent portfolio in 2026 needs to ask a narrower question than portfolio size: how many of the claims describe a specific technical mechanism, similar to Aon's classifier arrangement, versus a functional outcome, similar to State Farm's model-generation patent. The answer changes the multiple. It also changes the appropriate representation and warranty language: a seller's warranty that patents are "valid and enforceable" carries materially different risk today than it did before February 2026, and buyers' counsel should be pricing that gap into indemnification caps and escrow terms rather than treating patent validity as boilerplate.

This is a concrete input for actuaries doing reserving or pricing consulting work inside an M&A engagement, not just for patent counsel. Purchase price allocation exercises that capitalize acquired AI patents as identifiable intangible assets under acquisition accounting inherit the same invalidation risk, and that risk should inform the impairment testing assumptions built into post-close financial reporting. A patent portfolio capitalized at a multiple that assumed durable enforceability, then tested for impairment after a district court grants a Section 101 motion to dismiss against a similar claim, is exactly the kind of contingent event actuaries supporting financial reporting engagements should be flagging before it becomes a surprise write-down.

Why This Belongs on the ERM Risk Register

Patent invalidation exposure is a contingent legal risk, and carriers with large patent books should be tracking it the same way they track cyber liability or reserve adequacy risk: as a named item on the enterprise risk register with an owner, a trigger, and a monitoring cadence. The trigger is straightforward to define now that a pattern exists. Four Federal Circuit rulings since April 2025 (Recentive, GoTV, Innovaport, and RPI v. Amazon) have invalidated AI and software patents sharing the "generic technique applied to a known problem" structure, and a carrier's own Tier 1 patents, the ones built on that same structure, sit one motion to dismiss away from the same outcome in active litigation. For carriers that have licensed patents to third parties or that rely on patent deterrence as part of a competitive strategy, that is a revenue and moat risk, not merely a legal cost.

Enterprise risk actuaries do not need to become patent litigators to incorporate this. The useful discipline is treating patent portfolio composition, what share of a carrier's AI patents fall into the generic Tier 1 pattern versus the more defensible categories, as a measurable input alongside the other contingent liabilities already on the register. A carrier with a patent book concentrated in claims-triage and pricing-automation patents drafted before 2023 carries materially more of this exposure than one whose recent filings emphasize specific classifier arrangements, defined data pipelines, or multi-agent coordination mechanisms, because the drafting choices made before the 2025 reset simply did not anticipate the standard now being applied.

What to Watch Through the Rest of 2026

Three developments will determine how much of the exposure identified here becomes concrete. Federal Circuit follow-on cases will keep testing the boundaries of what counts as a technical improvement sufficient for eligibility, and each new opinion refines the line between Aon's surviving claim and State Farm's exposed one. District court motions to dismiss targeting insurer or insurtech AI patents specifically, rather than the streaming, retail, and voice-assistant patents decided so far, will be the first direct test of whether insurance claims fare better or worse than the pattern suggests. And the pace of vendor consolidation, Sixfold, Cytora, and the services companies covered in our EXL patent portfolio analysis, will keep signaling how much the market has already priced this risk into build-versus-buy decisions, regardless of what the courts do next.