From reviewing USAA’s patent filings in the USPTO PAIR database, the agentic claims describe architectures that go well beyond simple chatbot automation. These filings cover multi-step decision orchestration that mirrors how experienced underwriters actually evaluate complex risks: intake agents extract and normalize submission data, risk profiling agents build comprehensive assessments against underwriting guidelines, compliance agents check regulatory adherence, and a decision orchestrator aggregates everything to determine whether a case can proceed automatically or needs human escalation. Only three insurers have filed agentic AI patents at all, and USAA dominates the category. This piece examines what “agentic” means in the patent context, why the rest of the industry is not filing, and what the agentic patent gap signals about the next phase of insurance AI strategy.

Evident’s Insurance AI Patent Tracker, released in December 2025, established that State Farm (326 patents), USAA (218), and Allstate (136) control 77% of all insurer AI filings since 2014. But buried within that aggregate number is a subcategory that no trade press outlet has isolated: agentic AI patents. While generative AI patents surged from 4% to 31% of all filings between 2014 and October 2025, agentic patents remain vanishingly rare. Their scarcity makes them strategically significant. As carriers increasingly deploy multi-agent workflows in production, the gap between operational adoption and IP protection is widening into a vulnerability that will surface during M&A due diligence, vendor negotiations, and regulatory examinations.

What makes a patent “agentic” versus “generative”

The distinction between generative and agentic AI patents is architectural, not just terminological. Generative AI patents cover systems that produce outputs in response to prompts: text summaries, image enhancements, structured data extractions. The claims describe input-output pipelines where a model receives data, processes it, and returns a result. USAA’s own generative AI patent for clarifying aerial property damage imagery is a textbook example. After a hurricane or severe storm, the patented system uses generative models to enhance, clarify, and annotate satellite or drone imagery, improving the speed and accuracy of catastrophe claims assessments for military families whose insured properties may be thousands of miles from their current station.

Agentic patents describe something fundamentally different. The claims cover system-level designs with three defining characteristics:

Multi-agent coordination. Rather than a single model processing a single task, agentic patent claims describe architectures where multiple specialized AI agents operate in concert. Each agent handles a discrete function (document extraction, risk scoring, compliance verification, pricing calculation), and the patent covers how those agents communicate, share context, and hand off structured outputs to one another. The coordination mechanism itself is the patentable innovation, not the individual model capabilities.

Control mechanisms and guardrails. Agentic patent claims typically include explicit governance layers that constrain what actions agents can take autonomously versus what requires human escalation. These control mechanisms distinguish agentic architectures from simple automation. The patents describe decision boundaries: under what conditions an agent can approve a claim or bind a policy without human review, and what triggers an escalation. This is the architectural feature that separates a claims chatbot from an autonomous claims processing system.

Continuous feedback loops. The third defining feature is adaptive behavior. Agentic patent claims describe systems where agent performance is monitored, outcomes are fed back into the coordination layer, and agent behavior adjusts based on results. This is distinct from traditional machine learning retraining cycles (which happen periodically and offline) because the feedback operates within the workflow itself, allowing the system to refine its routing, prioritization, and escalation decisions in near real-time.

Patterns we’ve seen across the limited agentic filings suggest these three elements form the minimum threshold. A patent that describes only multi-agent coordination without feedback loops or control mechanisms would likely be classified as an automation patent rather than an agentic one. The combination is what creates the architectural novelty that patent examiners look for when evaluating claims under Section 101 eligibility standards.

USAA’s agentic patent portfolio: what the claims actually describe

USAA leads the agentic patent category among insurers, and its filings reveal a coherent strategy focused on two operational domains: autonomous claims workflows and adaptive underwriting decision chains.

Autonomous claims workflows

USAA’s agentic claims patents describe end-to-end processing architectures where specialized agents handle the full claims lifecycle without requiring human intervention at each step. The typical architecture in these filings involves an intake agent that ingests and classifies incoming claim data (photos, descriptions, police reports, medical records); a validation agent that cross-references the claim against policy terms, coverage limits, and prior claims history; an assessment agent that evaluates damage severity using both structured data and unstructured inputs; and a resolution agent that calculates the payout, generates the settlement offer, and triggers payment processing.

What makes these claims agentic rather than simply automated is the orchestration layer that sits above the individual agents. This layer manages sequencing, handles exceptions, and implements the decision boundaries that determine when the system resolves a claim autonomously versus when it routes to a human adjuster. The feedback mechanism tracks resolution outcomes (customer acceptance rates, reopened claims, litigation frequency) and adjusts the orchestration logic accordingly. Over time, the system learns which claim types can be safely resolved without human review and which require escalation, effectively calibrating its own autonomy level.

For a carrier serving active-duty military families who may file claims from overseas deployment zones with limited ability to participate in traditional adjusting processes, the operational value of a system that can process straightforward claims autonomously is significant. The patent protection ensures that the specific architecture USAA developed to handle this use case cannot be replicated by competitors without licensing.

Adaptive underwriting decision chains

USAA’s underwriting-focused agentic patents describe multi-step evaluation workflows where the system mirrors the sequential judgment process that experienced underwriters use when assessing complex risks. Rather than feeding all available data into a single model and extracting a risk score, the patented architecture breaks the underwriting evaluation into discrete stages, each handled by a specialized agent:

  • A submission intake agent ingests and normalizes application data from multiple formats (PDFs, spreadsheets, structured API feeds), resolving inconsistencies and flagging missing fields
  • A risk profiling agent builds comprehensive risk assessments by cross-referencing the normalized data against underwriting guidelines, historical loss data, and external risk indicators
  • A pricing and product agent structures the policy terms and calculates the premium based on the risk profile, available capacity, and competitive positioning
  • A compliance agent reviews the proposed policy for regulatory adherence, checking rate filing requirements, coverage mandate compliance, and disclosure obligations
  • A decision orchestrator aggregates the outputs from all upstream agents to determine whether the case can be approved automatically or requires human underwriter review

The key patentable feature is the continuous feedback loop between the decision orchestrator and the individual agents. When the orchestrator routes a case to a human underwriter and that underwriter overrides the system’s preliminary recommendation, the override rationale flows back to the relevant agents, refining their future assessments. This creates an adaptive system that improves its autonomous decision-making over time without requiring full model retraining cycles. The feedback mechanism is what distinguishes the architecture from a conventional rules-based underwriting workflow with AI-assisted steps.

Why most carriers are not filing agentic patents

The scarcity of agentic AI patents across the insurance industry reflects a deliberate strategic calculus rather than a lack of technical capability. Multiple factors explain why most carriers that are actively deploying agentic systems in production are not seeking patent protection for those deployments.

The trade secret preference

The most significant factor is that most carriers prefer trade secret protection for their AI architectures. Trade secrets provide immediate protection with no application process, no public disclosure requirement, and no expiration date as long as confidentiality is maintained. For AI systems whose competitive value depends on specific training data, feature engineering, hyperparameter configurations, and orchestration logic, trade secret protection avoids the fundamental tension inherent in patents: you must publicly disclose your invention to receive legal protection for it.

The strategic calculus has shifted since the Defend Trade Secrets Act (DTSA) of 2016 created a federal cause of action for trade secret misappropriation. Trade secret litigation filings increased 25% within the first year of the DTSA’s passage, and over 1,200 trade secret cases were filed in U.S. courts in 2025 alone, according to the Berkeley Technology Law Journal. The federal framework gave carriers confidence that trade secret protection is enforceable, reducing the perceived need for patents.

For internal AI systems that are never exposed to customers or competitors, trade secrets are often the pragmatic choice. An insurer’s proprietary claims triage algorithm, for instance, operates entirely within the company’s infrastructure. No competitor can reverse-engineer it from the outside. No customer interaction reveals the underlying architecture. In these cases, the disclosure requirement of a patent creates risk without corresponding benefit.

Section 101 eligibility uncertainty

The legal landscape for AI patent eligibility also discourages filing. The Federal Circuit’s April 2025 decision in Recentive Analytics, Inc. v. Fox Corp. established that patents which merely apply known machine learning methods within a new data environment do not survive Section 101 scrutiny. The Supreme Court’s subsequent denial of certiorari in December 2025 cemented this as binding precedent.

For carriers considering agentic AI patent filings, the Recentive Analytics framework creates a specific challenge. If the patent claims simply describe “apply multi-agent coordination to insurance underwriting,” they will almost certainly be rejected under Section 101 as an abstract idea implemented on a generic computer. To survive, the claims must articulate a concrete technical improvement in how the agents coordinate, how the feedback loops operate, or how the control mechanisms govern autonomous decision-making. This level of specificity requires significant patent drafting expertise and deep technical documentation of the system architecture.

The November 2025 USPTO guidance and the Ex Parte Desjardins decision opened a clearer pathway for AI patents that demonstrate genuine technological improvements. But the gap between the USPTO’s more permissive approach and the Federal Circuit’s stricter standards creates uncertainty. Carriers and their patent counsel must draft claims that satisfy both the examiner and the courts, which means investing in detailed technical specifications that many insurers have not historically produced for their AI systems. Our Section 101 analysis covers this tension in detail.

Vendor-developed architectures

A growing share of agentic AI deployments at carriers runs on vendor platforms rather than internally developed systems. When a carrier deploys Palantir’s Foundry platform with its Agent Studio for orchestrating multi-agent workflows, or uses Microsoft’s Copilot Studio to build agentic customer service systems, the patentable innovation often belongs to the vendor rather than the carrier. The carrier contributes domain expertise, training data, and workflow requirements, but the underlying multi-agent coordination technology is the vendor’s intellectual property.

AIG’s March 2026 partnership with McGill and Partners illustrates this dynamic. AIG deployed agentic AI using Palantir’s Foundry platform across a specialty insurance portfolio worth up to $1.6 billion in gross premiums written. The orchestration layer that coordinates AI agents to drive underwriting decisions was built on Palantir’s infrastructure. AIG CEO Peter Zaffino described the agents as “companions that operate with our teams,” providing real-time information and historical case context. But the platform architecture itself is Palantir’s IP, not AIG’s. We covered the full deployment in our analysis of AIG’s agentic underwriting architecture.

This vendor dependency creates a structural gap. The carrier uses the agentic system operationally but does not own the IP that makes it work. In an acquisition scenario, the buyer would acquire the carrier’s book of business and its vendor contracts, but not any proprietary agentic AI technology. USAA’s decision to develop and patent its own agentic architectures reflects a different strategic choice: retaining IP ownership even at higher development costs.

The trade secret trap: independent invention and M&A risk

Carriers that rely exclusively on trade secret protection for their agentic AI systems face risks that patents would mitigate. The most significant is the independent invention problem.

Unlike patents, trade secrets do not provide exclusive rights against independent development. If two carriers independently develop similar multi-agent claims processing architectures, neither can prevent the other from using its system as long as neither misappropriated the other’s proprietary information. This means trade secret protection does not create the competitive moat that carriers assume it does. A competitor can study publicly available information about agentic AI architectures (academic papers, vendor documentation, conference presentations) and build a functionally equivalent system without any legal liability.

The M&A implications are more immediate. When a carrier is acquired, the buyer’s due diligence team will assess the target’s AI capabilities as part of the technology valuation. Patents provide clear, verifiable evidence of proprietary technology. Trade secrets are harder to value because their existence depends on confidentiality measures that can be difficult to audit. Did the carrier require all AI team members to sign non-disclosure agreements? Did it implement access controls on its model architectures? Did former employees who joined competitors take knowledge with them? These questions create valuation uncertainty that can reduce the premium a buyer is willing to pay for the carrier’s technology assets.

Vendor negotiations present a parallel challenge. A carrier negotiating with a platform vendor like Palantir, Guidewire, or Verisk has more leverage if it holds patents on proprietary AI methods. The carrier can demonstrate that it brings genuine technical innovation to the partnership, not just data and domain expertise. Without patents, the carrier’s negotiating position is weaker because the vendor can argue that the carrier’s contribution is commodity-level data science applied to the vendor’s proprietary platform.

How agentic patents intersect with NAIC transparency requirements

The connection between agentic AI patents and regulatory compliance is underappreciated. The NAIC’s Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted in 2023 and now enacted in approximately 24 states, requires carriers to implement documented AI governance programs that prioritize transparency, fairness, and accountability. The bulletin explicitly states that “controls and processes in the AI program should be reflective of, and commensurate with, insurers’ assessment of the degree and nature of risk posed to consumers by the AI systems.”

For agentic AI systems that make autonomous decisions about claims and underwriting, the transparency requirement creates a specific challenge. Regulators want to understand how the system reaches its decisions: what data inputs it considers, how different factors are weighted, and what triggers human escalation. The NAIC’s multistate pilot of an AI Evaluation Tool, running from January through September 2026 across twelve participating states, is designed to give regulators a structured framework for examining exactly these questions during market conduct reviews.

Here is where patent strategy and regulatory strategy converge. A carrier that has patented its agentic AI architecture has, by definition, produced a detailed public description of how the system works. Patent claims must describe the method with sufficient specificity for a person skilled in the art to reproduce it. This level of documentation maps directly onto what regulators need when evaluating AI systems: a clear description of the decision-making architecture, the control mechanisms, the feedback loops, and the escalation criteria.

Carriers that protect their agentic systems exclusively through trade secrets face a dilemma when regulators request transparency. Providing detailed architectural documentation to regulators risks degrading the trade secret protection. If the documentation becomes part of a public regulatory record, the confidentiality that trade secret law requires may be compromised. Carriers must balance the regulatory obligation to explain their AI systems against the legal requirement to maintain secrecy. Patents eliminate this tension: the architecture is already public, and the carrier’s legal protection comes from the patent grant rather than from confidentiality.

This dynamic is particularly relevant under the NAIC’s anticipated transition from the current Model Bulletin (which is non-binding guidance) to an enforceable model law. As we analyzed in our coverage of the 33 comment letters on the model law transition, the enforcement provisions under consideration would give regulators explicit authority to examine AI decision-making systems during market conduct reviews. Carriers with patented architectures will find compliance more straightforward than those relying on trade secrets.

The adoption-versus-IP paradox: agentic deployment is outpacing patent activity

The most striking feature of the current landscape is the divergence between agentic AI adoption and agentic AI patent activity. Insurance AI deployments grew 87% year-over-year according to Evident’s Q4 2025 data, with agentic AI accounting for 21% of publicized deployments in that quarter. Of those agentic deployments, 56% focused on claims management and another significant share targeted underwriting and customer engagement.

Metric Value Source
Year-over-year AI deployment growth 87% Evident Q4 2025
Share of Q4 2025 deployments that were agentic 21% Evident Q4 2025
Agentic deployments focused on claims 56% Evident Q4 2025
Insurers planning agentic production systems by end of 2026 22% Industry surveys
Insurers with agentic AI patents 3 Evident Patent Tracker
Insurers reporting tangible AI business benefits 40% Evident Q4 2025

The numbers tell the story. Twenty-two percent of insurers plan to have agentic AI in production by the end of 2026, yet only three have filed agentic patents. Forty percent of insurers report tangible business benefits from AI implementations, with 77% of those benefits linked to productivity gains. The technology is delivering measurable value, but the IP protection is not keeping pace.

This gap represents a window. Carriers that file agentic patents now, while the category is nearly empty, can establish prior art and claim positions that will constrain latecomers. The early filers will define what counts as “obvious” versus “novel” in agentic insurance AI, setting the bar for every subsequent filing. The analogy is to the early 2010s, when a handful of carriers filed the first telematics patents and subsequently controlled the IP landscape for usage-based insurance for years. The agentic AI patent window is open now, but it will not stay open indefinitely.

Real-world benchmarks: what agentic systems actually deliver

The operational results from early agentic deployments demonstrate why the technology is attracting attention faster than the patent filings can track. Several data points from the past two quarters illustrate the scale of impact.

Hiscox compressed its London Market specialty quote cycle time by 99.4%, reducing turnaround from three days to approximately three minutes while preserving underwriter control over final pricing decisions. The system uses agentic workflows to handle submission intake, risk assessment, and preliminary pricing autonomously, escalating only the final pricing decision to human underwriters.

AIG’s deployment through its McGill and Partners collaboration processes submissions across a $1.6 billion specialty portfolio using Palantir’s Foundry platform. AIG’s Lexington Insurance subsidiary surpassed 370,000 submissions processed in 2025, targeting 500,000 by 2030 without proportional staffing increases. The agentic orchestration layer coordinates multiple AI agents across the front-to-back workflow, from intake through risk assessment and binding.

Allianz’s Project Nemo achieved an 80% reduction in food spoilage claims processing time using agentic workflows for first-notice-of-loss through settlement. Swiss Re’s Wysa Assure program demonstrated a 31% reduction in depression rates among policyholders using AI-powered mental health support, with projected 33% drops in related claims.

These results explain the adoption acceleration. They also explain why the patent gap matters: the carriers deploying these systems are building operational advantages that, without patent protection, are vulnerable to replication by any competitor willing to invest in similar vendor relationships and technical talent.

The Section 101 challenge for agentic patent drafting

Carriers considering agentic AI patent filings must navigate the post-Recentive Analytics eligibility framework carefully. The Federal Circuit made clear in April 2025 that patents claiming the application of known AI methods to a new domain do not clear the Section 101 bar. For agentic insurance patents, this means the claims cannot simply describe “use multi-agent coordination for underwriting” and expect to survive scrutiny.

Successful agentic patent claims need to articulate what is technically novel about the specific coordination mechanism, feedback loop architecture, or control system. The November 2025 USPTO guidance, incorporating the Ex Parte Desjardins decision, provides a pathway: if the claims demonstrate a concrete improvement in the functioning of the AI system itself (not just an improvement in the business process the AI supports), they can overcome Section 101 challenges.

For insurance agentic systems, this means patent claims should focus on:

  • Novel coordination protocols that improve how agents share context, resolve conflicts between competing assessments, or handle cascading failures when one agent produces unreliable outputs
  • Feedback architectures that demonstrate measurably improved system performance over time, with specific metrics tied to the adaptive mechanisms described in the claims
  • Control mechanisms that implement governance constraints in technically novel ways, such as dynamic autonomy thresholds that adjust based on real-time confidence scores across the agent ensemble
  • Integration innovations that solve specific technical problems in combining structured and unstructured insurance data across the multi-agent pipeline

USAA’s existing filings appear to satisfy these requirements. The multi-agent architectures described in their claims include specific technical details about coordination protocols and feedback mechanisms, not just high-level descriptions of applying AI to insurance processes. This level of specificity is what distinguishes a survivable patent from one that will be invalidated under Recentive Analytics.

Strategic implications for carriers and actuaries

The agentic patent landscape carries implications for three constituencies: carrier leadership, actuarial practitioners, and the vendors building the platforms.

For carrier leadership, the IP strategy question is urgent. Every month that a carrier deploys agentic AI systems without filing patents is a month where competitors and vendors can develop overlapping claims. The trade-secret-only approach may feel safer in the short term, but it creates long-term vulnerabilities in M&A, vendor negotiations, and regulatory compliance. Carriers should audit their existing agentic deployments and identify architectures that contain genuinely novel coordination mechanisms, feedback loops, or control systems. Those components are candidates for patent filing, even if the broader system architecture remains protected as a trade secret. The hybrid approach (patenting novel architectural elements while protecting implementation details as trade secrets) is the strategy most IP attorneys recommend for AI systems.

For actuarial practitioners, the agentic patent landscape matters because it signals where autonomous decision-making is most advanced. The patent claims describe specific use cases where AI agents make actuarial judgments: risk classification, reserve estimation, pricing optimization, compliance verification. These are functions that have traditionally been performed by credentialed actuaries operating under professional standards. As agentic systems take over more of these functions, the profession’s role shifts toward designing the governance frameworks, calibrating the decision boundaries, and validating the feedback mechanisms that patent claims describe. Understanding what these patents actually claim helps actuaries anticipate how their roles will evolve.

For platform vendors, the carrier patent gap is an opportunity. Palantir, Guidewire, Microsoft, and other platform providers can position themselves as the IP layer in the stack: if carriers are not patenting their own agentic architectures, the vendor patents become the controlling IP in the space. This strengthens vendor lock-in and increases switching costs for carriers. Carriers that want to maintain strategic flexibility should view their own patent filings as a counterweight to growing vendor IP portfolios.

What to watch in the next 12 months

Evident expects agentic patent activity to increase in 2026, with a growing focus on system-level designs, multi-agent coordination, control mechanisms, and continuous feedback loops. Several specific developments will shape the landscape:

The NAIC AI Evaluation Tool pilot results (due September 2026) will reveal how regulators approach agentic systems during market conduct examinations. If regulators demand the level of architectural transparency that patent filings provide, the regulatory advantage of patenting over trade secrets will become concrete rather than theoretical.

The model law transition, if the NAIC moves from the current non-binding bulletin to enforceable legislation, will create compliance obligations that favor documented, transparent AI architectures. Carriers that have already filed patents will have a compliance head start.

AIG’s Palantir deployment results will establish benchmarks for agentic underwriting at scale. Whether AIG chooses to file its own agentic patents (distinct from the Palantir platform patents) will signal whether major carriers are reconsidering the trade-secret-only approach.

USAA’s continued filing activity will expand the prior art base in the agentic category, progressively narrowing the design space available to latecomers. Each new USAA filing defines what is “known” in the field and raises the novelty threshold for subsequent applications.

Further Reading

Sources