Insurer AI patent filings look like a slowdown in aggregate. The 166 patents filed by 30 major North American and European insurers between January 2023 and October 2025 leave cumulative volume roughly 30% below the 2020 peak. Inside that aggregate the composition moved sharply: generative AI grew from 4% of insurer AI filings in 2014 to 31% as of October 2025, according to Evident's Insurance AI Patent Tracker.

The tracker was read as a carrier ranking. The more useful reading is where inside the portfolio those GenAI filings landed, and where they did not.

Key Takeaways

  • 31% of insurer AI filings are now generative, up from 4% in 2014, but the GenAI layer clusters in customer service and claims rather than the underwriting and pricing that carriers name as their primary investment destination.
  • State Farm (326), USAA (218), and Allstate (136) hold 77% of the tracked portfolio. The roughly 27 other carriers average fewer than eight filings each across twelve years.
  • P&C carriers hold 89% of all insurer AI patents, which is a Section 101 artifact as much as a market-size one: sensor and imagery claims clear eligibility that statistical inference on administrative data does not.
  • Only three insurers have filed agentic AI patents, while agentic systems accounted for 21% of publicized insurer AI deployments in Q4 2025.
  • Cumulative volume sits roughly 30% below the 2020 peak, with vendor-platform deployment and trade secret preference explaining more of that gap than prosecution lag does.

What the Tracker Counts

The aggregate decline and the composition shift are two different facts, and only one of them is about how much AI carriers are building.

State Farm, USAA, and Allstate hold 680 patents between them, roughly 77% of a tracked universe implying about 883 total insurer AI filings since 2014. The concentration is not a recent development. It reflects sustained prosecution programs that the rest of the market never built.

Carrier AI Patent Filings (Since 2014) Approximate Share
State Farm 326 ~37%
USAA 218 ~25%
Allstate 136 ~15%
All other tracked carriers (27) ~203 ~23%
Total tracked ~883 100%

The 89% P&C share is structural rather than incidental. Telematics processing, sensor-based risk monitoring, usage-based rating, aerial and satellite imagery for property damage, and IoT underwriting triggers all describe concrete physical sensing methods that satisfy the technical improvement requirement under 35 U.S.C. Section 101. Life and health applications, including mortality prediction, morbidity trend, and medical cost projection, run on statistical inference against administrative data. That is the framing the Federal Circuit rejected in Recentive Analytics, Inc. v. Fox Corp. in April 2025, and which became final when the Supreme Court denied certiorari in December 2025.

Outside the Big Three, the roughly 27 tracked carriers average fewer than eight filings across twelve years. That is opportunistic filing around discrete innovations, not a patent program.

Where the GenAI Filings Landed

The 31% concentrates in customer interaction, and the reason is a disclosure calculation rather than a capability gap.

Evident's tracker shows GenAI filings clustered in two places: customer service applications such as chatbots, inbound query classification, and policy explanation tools; and claims processing, including FNOL summarization, damage description parsing, and settlement communication generation. Carrier investor materials from 2023 through early 2026 name something else as the priority. Travelers deployed Anthropic to 10,000 engineers across actuarial and engineering workflows. AIG's Palantir partnership targets underwriting automation across a $1.6 billion specialty portfolio.

Both things are true once the disclosure cost is priced. A patent on personalized renewal communications claims language model inference conditioned on structured insurance data, which is a concrete output-generation method with a reasonable path through examination. A patent on generating better initial loss reserve estimates by synthesizing unstructured claims notes with structured exposure data is, in effect, a claim over statistical actuarial judgment. It carries higher Section 101 rejection risk, and it publishes the reserving methodology to every competitor who reads the application.

The consequence lands on validation documentation. The NAIC multistate AI evaluation pilot runs through September 2026 across twelve participating states, and the NAIC Model Bulletin on AI is enacted in roughly 24. A carrier whose claims AI is patented has already written the specification an examiner asks for: inputs, outputs, decision pathways, and the boundary between autonomous and human resolution. A pricing model held as a trade secret has no equivalent artifact. Its validation record stands on internal methodology papers, and producing architectural detail for a public regulatory file puts the protection itself at risk.

The Agentic Layer Is Forming With Almost No Public Record

The distance between what carriers run and what they document is widest exactly where the systems are most autonomous.

Three insurers have filed agentic AI patents at all. USAA leads, with multi-agent coordination claims covering aerial imagery property damage assessment and adaptive underwriting decision chains with continuous feedback loops. Evident's separate Q4 2025 use case report put agentic systems at 21% of publicized insurer AI deployments, with claims management accounting for 37% of those projects and five insurers actively testing agentic AI in claims.

The examination environment is no longer the constraint. The USPTO's November 2025 guidance revision rescinded substantial portions of the February 2024 AI guidance and advanced Ex Parte Desjardins as a precedential PTAB decision, giving claims that improve the functioning of an AI system a cleaner path than they had 18 months ago. Agentic coordination protocols, control mechanisms, and feedback loop architectures are precisely the claim type that reset favors. The Section 101 reset analysis covers the claim structures the Desjardins framework supports.

First filings in a subcategory set what later examiners treat as obvious. Carriers that did not file telematics patents in 2013 and 2014 are still designing around State Farm's positions a decade later, and the detailed read of USAA's agentic portfolio shows the same pattern forming again with one dominant filer.

The practical effect is that the tracker cannot be read as a capability ranking, and the direction of its error is not random. It is thinnest exactly where the systems are most consequential: underwriting, pricing, and the agentic tier already running in production at carriers with no filings at all. An actuary assessing model risk from the public record gets rich documentation of renewal letters and almost none of the machinery making decisions.

Further Reading

Sources