From tracking USPTO filings across the top 20 carriers for the past 18 months, the shift from traditional ML to generative and agentic patent claims marks the clearest IP inflection point since telematics patents peaked in 2019. Between 2014 and October 2025, generative AI patents climbed from 4% to 31% of all insurer AI filings, according to Evident’s Insurance AI Patent Tracker. That is not a gradual drift. It is a structural rebalancing of what carriers consider worth protecting, and it carries direct implications for competitive positioning, freedom-to-operate risk, and the Section 101 eligibility landscape that governs whether these filings survive legal challenge.

The trade press covered the three-insurer concentration story (State Farm 326 patents, USAA 218, Allstate 136) and the USPTO guidance changes in separate reporting cycles. This analysis connects a layer that has gone largely unexamined: the patent type transition from classical machine learning to foundation-model architectures, and what it signals about how carriers will compete on AI over the next three to five years.

The data: mapping the generative AI patent surge

Evident, the AI benchmarking platform for financial services, tracks AI patent filings by 30 major insurers across North America and Europe. The dataset covers all AI-related patents filed since 2014 and classifies them by technology type (traditional ML, generative AI, agentic AI) and functional area (claims, underwriting, customer service, risk modeling and pricing).

The headline numbers tell a clear story. Since January 2023, these 30 insurers have filed 166 AI patents. That pace remains roughly 30% below the 2020 peak, when insurer AI patent activity was at its highest. But the composition of those filings has changed dramatically.

Patent Category Share of Filings (Pre-2023) Share of Filings (2023-2025) Primary Applications
Traditional ML ~96% ~66% Claims triage, risk scoring, fraud detection
Generative AI ~4% ~31% Customer service, claims automation, document generation
Agentic AI <1% ~3% Multi-agent underwriting, adaptive feedback systems

P&C carriers account for 89% of all insurer AI patents filed since 2014. That structural dominance holds across all three technology categories. Life and health carriers have filed comparatively few patents, a pattern that reflects P&C’s natural overlap with sensor data, image recognition, and real-time pricing models that produce more patentable claims than actuarial model refinements.

“Patents offer a rare window into where insurers are placing their biggest bets on AI,” said Alexandra Mousavizadeh, cofounder and CEO of Evident. “This data shows that innovation is overwhelmingly being driven by a handful of U.S. firms, especially in P&C.”

Where GenAI filings concentrate: claims and customer service lead

The functional distribution of generative AI patents reveals where carriers see the fastest return on IP investment. Claims and underwriting together account for over 300 AI patents across the decade, more than twice the volume of the next most common focus areas. But within the generative AI subset, customer service and claims automation dominate, while underwriting applications are just beginning to emerge.

This concentration pattern makes operational sense. Customer service and claims processing are the two highest-volume, highest-cost functions in most carrier operations. Generative models can reduce cost per transaction quickly in these areas through automated first-notice-of-loss intake, policyholder communication drafting, claims summarization for adjusters, and chatbot-driven service interactions. These applications produce measurable savings within months of deployment, which explains why carriers are protecting the specific implementations through patent filings.

Risk modeling and pricing patents, by contrast, remain anchored to traditional machine learning. This is not because carriers have neglected GenAI for pricing; rather, pricing model innovations tend to rely on well-known statistical techniques applied to proprietary data combinations, making them harder to patent and easier to protect as trade secrets. The competitive moat in pricing comes from the data itself, not the algorithm, and data is better protected through confidentiality than through public patent disclosure.

Patent examples by application area

The Evident tracker highlights specific filings that illustrate the functional focus:

  • USAA: Generative AI to clarify and enhance aerial imagery for property damage assessment. After hurricanes or severe convective storms, adjusters historically relied on manual inspection or raw satellite imagery. USAA’s patented approach uses generative models to annotate and enhance aerial views, accelerating the catastrophe claims cycle for a carrier whose military-family policyholders are often stationed far from their insured properties.
  • State Farm: Machine learning for claims triage and autonomous vehicle fault analysis. State Farm’s 326-patent portfolio, the largest in the industry, clusters around systems that automatically sort incoming claims by complexity, route straightforward cases to straight-through processing, and analyze sensor data from semi-autonomous vehicles to determine collision fault.
  • Allstate: An in-vehicle AI assistant that automates elements of the claims process and offers behavior-based discounts in real time. This system uses vehicle sensors to monitor driving behavior, categorize the driver’s risk profile, and dynamically adjust pricing signals, extending the Drivewise usage-based insurance program into AI-driven continuous behavioral monitoring.
  • Liberty Mutual: Generative AI for engineering release notes, a more specialized application that reflects the diversity of GenAI use cases beyond the customer-facing applications that dominate the headlines.
  • Swiss Re: Predictive analytics for medical data and anomaly detection systems, representing the reinsurance side of the AI patent landscape.

The three-carrier concentration: a strategic moat in formation

State Farm (326 patents), USAA (218), and Allstate (136) together hold 77% of all insurer AI patents filed since 2014. We covered the implications of this concentration in detail in our analysis of the Evident patent tracker release. But the generative AI type shift adds a new dimension to the concentration story.

Each carrier’s portfolio reflects a different strategic bet on where GenAI creates protectable value:

Carrier Total AI Patents GenAI Patent Focus Agentic AI Activity
State Farm 326 Claims automation, autonomous vehicle fault analysis Limited
USAA 218 Aerial imagery enhancement, property damage assessment Leads all insurers; multi-agent coordination filings
Allstate 136 In-vehicle AI assistant, telematics-driven pricing Interpretable AI for underwriting decisions

The distinction that matters for competitive positioning is not the total count but what type of AI each carrier is protecting. State Farm’s portfolio is heavily weighted toward operational automation of existing workflows. USAA is the outlier filing on architecturally novel systems, particularly in agentic AI. Allstate’s filings emphasize the telematics-to-pricing pipeline. These are three different answers to the same question: where does AI generate enough proprietary value to justify the public disclosure that a patent requires?

The remaining 27 carriers in the Evident tracker share just 23% of total filings. Swiss Re, MassMutual, Liberty Mutual, and Zurich each hold specialized portfolios, but none approaches the volume of the top three. For carriers like Chubb, which holds zero AI patents despite announcing a multi-year digital transformation that will automate 85% of major underwriting and claims processes and cut headcount by roughly 20%, the strategic choice is explicit: build AI capabilities without creating public patent records, relying instead on trade secret protection and vendor platforms.

Agentic AI patents: the 2026 inflection point

If generative AI patents reflect where carriers are deploying today, agentic AI patents signal where they plan to compete tomorrow. As of the Evident tracker’s December 2025 release, only three insurers had filed agentic AI patents, with USAA leading the category. We analyzed this gap in depth in our coverage of why USAA leads and most carriers lag in agentic AI patents.

Evident forecasts that agentic patent activity will increase in 2026, with filings focused on system-level designs, multi-agent coordination architectures, and continuous feedback loop mechanisms. The high-impact areas for these filings are underwriting and claims handling, where the complexity of real-world workflows creates room for genuinely novel system designs that can survive Section 101 scrutiny.

The agentic AI market in insurance supports this forecast. The market is projected to grow from $5.76 billion in 2025 to $7.26 billion in 2026, a 26% growth rate (AI Journal). Twenty-two percent of insurers plan to have an agentic AI solution in production by year-end 2026. As these deployments move from pilot to production, they generate the operational architectures that become patentable claims.

What “agentic” means in patent claims

The distinction between generative and agentic AI is functionally clear but legally nuanced. Generative models produce outputs (text, images, structured data) in response to prompts. Agentic systems take autonomous action: they execute multi-step workflows, coordinate with other AI agents, make decisions within defined guardrails, and adjust their behavior based on outcomes without requiring human intervention at each step.

In insurance underwriting, the agentic architecture looks like this: an intake agent ingests and clarifies submission data; a risk profiling agent builds a comprehensive risk profile; a pricing and product agent structures the policy terms; a compliance agent checks regulatory adherence; and a decision orchestrator aggregates inputs to determine whether the case can be approved automatically or requires human escalation. Each agent is specialized, and the system-level coordination protocol is where the patentable innovation lies.

AIG’s Q1 2026 earnings call disclosed exactly this kind of deployment. In close partnership with Palantir and Anthropic, AIG has begun what CEO Peter Zaffino described as “the next phase of agentic AI,” using Palantir’s Foundry platform to expand an ontology that maps underwriting processes, workflows, and data relationships. This ontology, coupled with orchestration, enables the deployment of multiple AI agent teams that integrate with core systems. The early results are measurable: in Lexington middle market property, AIG Assist has delivered a 30% improvement in quoting more submissions, reduced time to quote by 55%, and increased binding of submissions by approximately 40%.

These production deployments at AIG, combined with USAA’s existing agentic patent filings, create the conditions for a surge in agentic patent activity. Carriers that have proven their systems work operationally now have both the architectural blueprints and the economic justification to file patent claims.

Build vs. buy: how patent strategy reflects platform choices

The generative AI patent data reveals a structural divide in how carriers approach AI implementation, and that divide directly affects who files patents and what they protect.

Build carriers develop proprietary AI workflows internally and patent the specific technical implementations. State Farm, USAA, and Allstate exemplify this approach. Their patent portfolios reflect innovations that emerged from internal R&D: specific claims triage algorithms, aerial imagery processing methods, telematics-to-pricing pipelines. The competitive moat is the patented method itself.

Buy carriers deploy vendor platforms and customize them for their operations. When a carrier uses Guidewire’s ClaimCenter with AI-assisted triage, or Verisk’s Synergy Studio for catastrophe modeling, the patentable innovation belongs to the vendor, not the carrier. The carrier’s competitive advantage comes from how well it configures and operates the vendor’s platform, not from novel technology that it invented.

Hybrid carriers combine vendor infrastructure with proprietary orchestration. AIG’s approach is the clearest example: Palantir provides the Foundry platform and ontology framework; Anthropic provides the Claude LLM backbone; and AIG has patented its own methods for document extraction, LLM traceability, and spreadsheet processing that sit on top of the vendor stack. This hybrid approach allows the carrier to claim IP on the integration and workflow layers while relying on vendor infrastructure for the foundation.

The patent data suggests this hybrid model may become the dominant strategy for large commercial carriers. Pure build requires the R&D investment that only the top three have sustained. Pure buy leaves no IP moat. The hybrid approach lets carriers file on what they uniquely contribute: the domain-specific orchestration, the proprietary training data pipelines, and the agentic coordination protocols that turn generic AI platforms into insurance-specific systems.

Travelers provides a notable data point. The carrier spent over $1.5 billion on IT in 2025, more than doubling its technology investment over eight years. Over 20,000 Travelers employees use AI tools regularly, and the company recently equipped 10,000 engineers and data scientists with Anthropic-powered AI assistants. Yet Travelers has filed relatively few AI patents, preferring to invest in operational capability rather than IP protection. Whether this leaves the carrier exposed to licensing demands from the top three patent holders remains an open question.

Section 101 eligibility: the legal filter on GenAI patents

The surge in generative AI patent filings runs directly into the legal uncertainty surrounding Section 101 patent eligibility. We covered the full framework in our analysis of the USPTO Section 101 reset, but the specific implications for GenAI-type patents deserve attention here.

The Federal Circuit’s April 2025 decision in Recentive Analytics, Inc. v. Fox Corp. established that “applying known machine learning methods within a new data environment” and “merely implementing abstract ideas using generic models on conventional infrastructure” do not constitute patent-eligible subject matter. The Supreme Court’s December 2025 cert denial let that precedent stand. For insurance AI patents, this creates a clear vulnerability: GenAI patent claims that describe using a large language model to perform a known insurance function (claims summarization, policyholder communication, document classification) without a specific technical improvement may not survive eligibility challenges.

However, the patent landscape has shifted in a countervailing direction under USPTO Director John A. Squires, who took office in September 2025. Three developments favor AI patent applicants:

  1. The Kim Memo (August 2025): Deputy Commissioner Charles Kim reminded examiners that the “mental process” rejection should not be applied too broadly to machine learning claims, since algorithms processing massive datasets cannot realistically be performed mentally.
  2. Ex Parte Desjardins (September 2025): A precedential Appeal Review Panel decision vacated a rejection of machine learning claims, establishing that improvements to how ML models function constitute patent-eligible subject matter. The panel cautioned against “categorically excluding AI innovations from patent protection.”
  3. Subject matter eligibility declarations (December 2025): Applicants can now submit objective evidence and expert testimony under Rule 132 to demonstrate technological improvements, strengthening arguments against Section 101 rejections.

For insurers filing generative AI patents in 2026, the practical guidance is to draft claims that emphasize specific technical improvements rather than broad functional descriptions. A patent claim covering “using a generative model for claims summarization” is vulnerable. A claim covering a specific method for integrating multi-source document extraction with chain-of-thought reasoning to generate auditable claims summaries with traceable source attribution is far more defensible.

Inventorship under the new guidance

The USPTO’s revised inventorship guidance, which rescinded the February 2024 framework in November 2025, adds another compliance layer. AI systems cannot be named as inventors or joint inventors; they are “merely tools for use by inventors who are natural persons.” For carriers deploying GenAI in their R&D processes, this means that patent applications must identify the human engineers who conceived the specific innovation, even when the AI system contributed significantly to the development process.

This is particularly relevant for the hybrid build-buy model. When AIG’s engineers build a proprietary orchestration layer on top of Anthropic’s Claude and Palantir’s Foundry, the human inventors are the AIG employees who designed the orchestration architecture, not the AI models that may have assisted in generating code or processing data. Getting this right at the application stage matters, because priority claims between applications now require at least one common natural person inventor.

What the type shift means for each dominant filer

The ML-to-GenAI transition creates different strategic imperatives for the three carriers that control 77% of the patent landscape.

State Farm: the largest portfolio faces an adaptation test

State Farm’s 326-patent portfolio is the deepest in traditional ML applications. Its claims triage and autonomous vehicle fault analysis patents are built on classical supervised learning, sensor data processing, and rules-based routing. The question for State Farm is whether this portfolio transitions naturally into generative and agentic territory or whether the carrier needs a fundamentally different R&D investment to compete in the next generation of AI patents.

The autonomous vehicle patents represent State Farm’s longest-term strategic bet, and they may age better than the claims triage patents. As connected and autonomous vehicles grow as a share of the insured fleet, the machine-to-insurer data pipeline that State Farm has patented becomes more commercially valuable. But claims triage using classical ML is increasingly commoditized through vendor platforms, which could erode the competitive moat that earlier patents established.

USAA: best positioned for the agentic wave

USAA’s 218-patent portfolio is the most forward-looking among the top three. Its lead in agentic AI patents, combined with generative AI filings for property damage assessment, positions USAA at the intersection of the two technology categories that Evident expects to dominate future filing activity. USAA’s multi-platform AI strategy (spanning Google Gemini, AWS Bedrock, Meta’s Llama, and internal models) means it is not locked into a single vendor stack, which preserves optionality for the kinds of multi-agent coordination architectures that generate patentable claims.

The risk for USAA is that its military-family focus limits the commercial applicability of some patents. Aerial imagery enhancement for catastrophe claims is broadly valuable, but patents designed around the specific operational constraints of serving deployed military members may have narrower defensive value in cross-licensing negotiations with carriers serving different markets.

Allstate: the telematics-to-GenAI bridge

Allstate’s 136-patent portfolio has a natural bridge to generative AI through its telematics and in-vehicle intelligence systems. The Drivewise program already collects continuous behavioral data; generative models can enhance how that data is interpreted, communicated to policyholders, and used in claims automation. Allstate’s patents on interpretable AI for underwriting decisions also align with the NAIC’s emerging regulatory framework, which may require carriers to explain AI-driven decisions to regulators and consumers.

Allstate’s strategic challenge is volume. At 136 patents, it holds roughly 40% of State Farm’s count and 62% of USAA’s. If the agentic AI patent surge materializes in 2026 and 2027, Allstate will need to increase its filing pace to maintain its position among the top three, or accept that its IP moat narrows relative to carriers that invest more aggressively in next-generation filings.

The patent gap beyond the top three

The 27 other carriers in the Evident dataset share roughly 203 patents across the full decade, averaging fewer than eight patents per carrier. For these companies, the generative AI type shift compounds an existing strategic gap.

Consider the carrier that begins filing generative AI patents in 2026. It faces a landscape where State Farm, USAA, and Allstate already hold issued claims on core applications. Filing on “using GenAI for claims automation” is likely to encounter prior art from one of the three dominant filers. The mid-market carrier must either find genuinely novel technical approaches, accept licensing exposure, or pursue the trade-secret-plus-vendor-platform strategy that Chubb has adopted.

This dynamic is particularly relevant for carriers evaluating build-vs.-buy decisions. We explored this in our analysis of the AIG vs. Quantiphi patent race, where the build-vs.-buy decision is complicated by overlapping IP claims from both carriers and vendors. A carrier building its own ML-driven claims triage system needs to consider whether its architecture infringes State Farm’s claims, while a carrier purchasing a vendor solution inherits whatever freedom-to-operate analysis the vendor has performed.

The vendor ecosystem provides some insulation. EXL holds 10 AI patents spanning document extraction, knowledge graphs, and a domain-specific insurance LLM. Verisk has integrated its analytics directly into Anthropic’s Claude through Model Context Protocol connectors. These vendor patents and platform integrations create a layer of IP protection that accrues to the vendor’s clients, but they do not give any single carrier a proprietary advantage over other clients of the same vendor.

What this means for actuaries and insurance professionals

The patent type shift from traditional ML to generative and agentic AI carries specific implications for actuarial practice and insurance operations.

Pricing model innovation remains largely unpatented. The Evident data confirms that risk modeling and pricing patents are a small share of total AI filings and remain anchored to traditional ML. Actuaries developing generative AI applications for pricing, whether through synthetic data generation, LLM-assisted rate filing analysis, or GenAI-enhanced loss development, are working in relatively open IP territory. The competitive moat for pricing innovation continues to reside in proprietary data and internal modeling expertise rather than in patent-protected methods.

Claims and underwriting actuaries face a more constrained landscape. With over 300 patents covering claims and underwriting AI, actuaries working on automation in these functional areas should be aware that the methods they help design may overlap with existing IP held by the top three filers. This does not mean every innovation is precluded, but it does mean that freedom-to-operate analysis should be part of the development process for internally built AI systems.

Regulatory compliance creates patent opportunities. The intersection of AI and regulatory requirements, particularly around model explainability, bias testing, and the NAIC AI risk taxonomy, creates a category of potentially patentable innovation that the current filing landscape has barely explored. Systems that can demonstrate compliance with Colorado’s AI Act or satisfy NAIC transparency requirements through novel technical methods may represent the next frontier of defensible AI patents.

The agentic transition will reshape workflow design. As carriers move from deploying individual AI models to orchestrating multi-agent systems, the actuarial workflow itself becomes a component of patentable system designs. An actuary who helps design the decision logic for an agentic underwriting pipeline is contributing to what may become a patent claim. Understanding this dynamic matters for career positioning and for the profession’s engagement with AI governance, a theme we explored in our analysis of the AI governance gap in actuarial practice.

Looking ahead: the patent landscape through 2027

Several forces will shape how the insurer patent landscape evolves over the next 18 months.

Agentic filings will accelerate. Evident’s forecast of increased agentic patent activity in 2026 is supported by the operational reality on the ground. AIG’s Palantir-Anthropic deployment, combined with production agentic systems at Travelers (20,000+ AI users), Chubb’s global claims AI mandate, and the broader insurer adoption curve (22% planning production agentic solutions by year-end 2026), will generate the system-level architectures that translate into patent filings.

Section 101 clarity will encourage filing. The Squires-era USPTO is signaling a more favorable posture toward AI patent applications. The Kim Memo, Ex Parte Desjardins, and the new eligibility declaration process reduce the unpredictability that discouraged some carriers from filing. If Section 101 rejection rates decline, the overall volume of insurer AI patent activity may recover toward the 2020 peak.

The UK precedent introduces a transatlantic dimension. The UK Supreme Court’s February 2026 ruling that an artificial neural network-based recommendation system was not excluded from patentability adds a European angle to patent strategy. Carriers operating in both U.S. and European markets may pursue parallel filings, expanding the geographic scope of their IP protection for GenAI and agentic systems.

Vendor-carrier IP boundaries will be tested. As hybrid build-buy models proliferate, disputes over who owns the IP generated when carrier engineers build proprietary layers on vendor platforms are likely. The Palantir-AIG-Anthropic configuration is the most visible example, but every carrier using a platform vendor for AI infrastructure will face the same question: where does the vendor’s IP end and the carrier’s begin?

Further Reading

Sources