State Farm, USAA, and Allstate Hold 77% of Insurer AI Patents: What the Concentration Means for the Rest of the Industry

Evident's Insurance AI Patent Tracker, released in December 2025, counts 326 AI patents filed by State Farm since 2014, 218 by USAA and 136 by Allstate. Those 680 patents are 77% of everything the 30 tracked insurers have filed.

The concentration is real. What it measures is filing behavior, and the categories where filing is rarest are the ones closest to ratemaking.

Key Takeaways

• Three carriers hold 680 of the tracked AI patents, 77% of the total, leaving roughly 203 across the other 27 insurers.
• Property and casualty carriers hold 89% of all insurer AI patents, a structural result of sensor, imagery and telematics data producing claims that survive eligibility review.
• Claims and underwriting together account for over 300 patents, more than twice the next largest category, while risk pricing patents are a small share.
• Generative AI grew from roughly 4% of filings to 31% by October 2025, yet total insurer AI patent activity peaked in 2020 and remains about 30% below that mark.
• Across all 30 insurers, 166 patents since January 2023 is about 5.5 per insurer over two years, and EXL, a services company rather than a carrier, holds 10.

The 77% Is Three Portfolios

Each of the three is protecting a different part of the business. State Farm's 326 patents cluster on machine learning for claims triage, sorting incoming claims by complexity and urgency, and on systems that read sensor data from autonomous and semi-autonomous vehicles to detect collisions and allocate fault. The portfolio runs from first notice of loss through settlement.

USAA's 218 are notable for position rather than volume. It leads the agentic category, which only three insurers have entered at all, and its generative filings include using generative models to enhance and annotate aerial imagery for property damage assessment after a storm.

Allstate's 136 sit at the intersection of AI and telematics, extending the Drivewise usage-based program: an in-vehicle assistant that automates parts of the claims process and adjusts behavior-based pricing signals in real time, plus filings on interpretable AI for underwriting decisions.

The 89% P&C share follows from the same structural fact. Telematics devices, IoT sensors, aerial imagery and vehicle diagnostics produce specific hardware integrations and signal-processing methods, which draft into claims that survive Section 101 review better than the purely algorithmic approaches typical of life and health AI. Short-tail feedback loops help too: an auto physical damage triage model can be evaluated in weeks, where a mortality model takes years.

The Record Is Blind to the Pricing Half

The category split inside the count matters more than the carrier split. Claims and underwriting together account for over 300 patents, more than twice the next largest category. Risk pricing patents exist but are a small share.

The reason is not that pricing innovation is rare. It is that pricing model innovations are hard to patent, because they typically apply well-known statistical techniques to new data, and easy to hold as trade secrets, because the value sits in the training data, the feature engineering and the parameterization rather than in a method a claim can describe.

That has a direct consequence for anyone reading competitive intent off these numbers. A carrier absent from the tracker may be investing heavily in the part of the AI stack that touches rate, and the absence would look identical either way. The 77% concentration is evidence about where three carriers chose to disclose, not about who is building better rating models.

The long tail carries the same caution. 166 patents since January 2023 across 30 insurers is roughly 5.5 each over two years, and much of the real filing sits outside the tracker's frame entirely: EXL, a services company, holds 10 AI patents spanning document extraction, knowledge graphs and a domain-specific insurance LLM. Vendor portfolios often exceed carrier portfolios at mid-market companies, and the tracker counts insurer-filed patents only.

Where the concentration does bite is freedom to operate. Machine learning claims triage is a capability dozens of carriers are building or buying, and a carrier that builds its own may find the architecture reads on claims already held. That risk is modulated by Recentive Analytics, which weakened broadly drafted AI claims, while narrow claims tied to specific implementations remain enforceable and are what sophisticated filers tend to hold.

The Decline Says More Than the Concentration

The finding that sits least comfortably with the moat reading is that insurer AI patent activity peaked in 2020 and remains roughly 30% below that level, while AI investment accelerated and generative filings went from about 4% to 31% of the total.

Trade secrets explain part of it. A patent requires public disclosure, handing competitors a blueprint alongside the exclusivity; a trade secret requires none and lasts as long as it holds. The Defend Trade Secrets Act of 2016 made that route more practical by creating a federal cause of action, and trade secret filings rose 25% within a year, with over 1,200 cases filed in US courts during 2025.

Eligibility explains more. The Federal Circuit's April 2025 decision in Recentive Analytics gave courts a template for invalidating generic apply-machine-learning-to-new-data claims, and the Supreme Court denied certiorari in December 2025. Broad AI claims now carry real invalidity risk, which discourages speculative filing without discouraging the underlying work.

Delegation explains the rest. When a carrier deploys AI-assisted triage inside a vendor platform, the patentable innovation belongs to the vendor. Open-source models absorb another slice: if an open-weight model handles claims summarization adequately, there is no patent-worthy alternative to fund.

Put together, the three carriers may hold 77% of the filings without holding 77% of the capability, and the gap is widening in the direction the tracker cannot see. None of the top three has launched a significant enforcement campaign against another carrier, so the portfolios currently function as deterrence and cross-licensing stock rather than as a toll on anyone's roadmap.