From tracking both carrier AI patent filings and insurance product filings across the top 20 P&C insurers, the divergence between internal AI investment and external AI coverage removal is accelerating faster than any prior technology-driven coverage cycle. In Q1 2026 alone, five of the six largest U.S. P&C carriers disclosed production AI deployments on earnings calls while simultaneously winning state regulatory approval to exclude AI-related damages from the commercial policies they sell. Chubb CEO Evan Greenberg publicly cited Anthropic’s Claude Mythos model as triggering a “new era of cyber-adjacent risk” during the same quarter his company filed AI exclusion endorsements in more than a dozen states. This is not a coverage adjustment. It is a structural repricing of who bears AI risk in the commercial insurance market, and actuaries sit at the center of both sides of the ledger.

This article maps the full paradox: what carriers are building internally, what they are removing from external coverage, how the litigation trajectory drove the exclusion wave, and what the resulting coverage gap means for enterprise risk managers and actuarial pricing in a peril class with essentially no loss history.

The Internal AI Build: What the Top Carriers Are Actually Deploying

The scale of internal AI deployment across major P&C carriers in 2026 is unprecedented. These are not pilot programs or innovation lab experiments. They are production systems processing real submissions, adjudicating real claims, and generating real underwriting decisions.

AIG deployed AIG Assist across its underwriting and claims operations, integrating Palantir Foundry with Claude to process commercial submissions at 88% accuracy rates on automated decisions. On its Q1 2026 earnings call, AIG described the platform as an “agentic AI underwriting machine” that processes 4 million data points per commercial risk. The system now handles initial submission triage, risk scoring, and coverage recommendation generation with human underwriters reviewing edge cases rather than routine binds.

Travelers rolled out Anthropic’s Claude to 10,000 employees, making it the largest disclosed carrier-to-foundation-model deployment in the industry. The $1.5 billion technology budget backing this rollout covers AI-assisted claims investigation, underwriting workflow automation, and policy language analysis. Travelers’ CEO described the deployment as “fundamental to how we operate,” not an efficiency experiment.

Allstate built ALLIE (Allstate Large Language Intelligence Ecosystem), a proprietary agentic AI stack that now handles customer engagement, direct policy sales in three states, and claims processing. Unlike carriers that licensed third-party models, Allstate chose to build its own platform, giving it control over training data, model governance, and deployment pace. ALLIE represents the most vertically integrated carrier AI investment disclosed publicly.

Chubb created a new executive role, Global Claims AI Mandate, under Jim Rampe. Greenberg disclosed on Chubb’s Q1 2026 earnings call that he now spends “much more time” on AI than even a year ago, warning that leaders who rely on second-hand briefings risk becoming “irrelevant.” The company is deploying AI across its global claims operation spanning 54 countries.

These are the same carriers filing to exclude AI risk from the policies they sell to their commercial clients. The contradiction is not subtle.

The External Exclusion Wave: What Carriers Are Removing From Coverage

While building AI internally, the industry moved with unusual coordination to strip AI-related liability from standard commercial policies. The exclusion wave operates through two channels: standardized ISO endorsements available to all carriers, and proprietary carrier-specific exclusion forms that go further.

ISO Endorsements: CG 40 47 and CG 40 48

Verisk’s ISO Core Lines Services made three new endorsement forms available to carriers nationwide effective January 1, 2026. The forms define generative artificial intelligence broadly as “a machine-based learning system or model trained on data with the ability to create content or responses, including but not limited to text, images, audio, video or code.”

CG 40 47 01 26 is the broadest form, excluding coverage under both Coverage A (bodily injury and property damage) and Coverage B (personal and advertising injury) for any loss “arising out of” generative AI. A manufacturer whose AI-generated product instructions cause injury, a retailer whose AI chatbot defames a competitor, a consulting firm whose AI-drafted report contains material errors: none would find coverage under a CGL policy endorsed with CG 40 47.

CG 40 48 01 26 is narrower, excluding only Coverage B (personal and advertising injury) while preserving Coverage A for bodily injury and property damage claims. This form targets the most visible near-term exposure: AI-generated content that infringes intellectual property, makes false statements, or invades privacy.

Because ISO forms underpin approximately 82% of U.S. commercial general liability policies, the adoption of these endorsements has systemic reach. Carriers do not need to draft their own exclusion language; they simply file to attach the ISO form to their existing CGL books.

W.R. Berkley’s Absolute AI Exclusion

W.R. Berkley went further than any ISO form with its proprietary Form PC 51380, an “absolute” AI exclusion that extends beyond CGL to Directors & Officers, Errors & Omissions, and Fiduciary Liability products. The endorsement eliminates coverage for any claim “based upon, arising out of, or attributable to any actual or alleged use, deployment, or development of Artificial Intelligence.”

The scope is sweeping. The exclusion covers content generation using AI, failure to identify AI-generated content, inadequate AI policies or training, breach of duties regarding AI deployment, products or services incorporating AI, chatbot or virtual customer service agent representations, AI-related statements and disclosures, violations of AI regulations, and regulatory requests to address AI impacts. For firms seeking coverage, Berkley requires affirmative warranties that AI outputs are reviewed by qualified personnel before professional reliance, that verification procedures are consistently followed and documented, and that all AI systems are inventoried and individually risk-assessed.

This is the most aggressive carrier-specific AI exclusion filed to date. It effectively tells commercial policyholders: if AI touched any part of your operations that led to a claim, we will not pay.

The Carrier Filing Wave and Regulatory Reception

Berkshire Hathaway, Chubb, and Travelers began submitting AI exclusion applications in fall 2025, with provisions taking effect as early as January 2026. State insurance regulators have approved more than 80% of these applications, with Florida, Connecticut, and Maryland processing approvals at the fastest pace. AIG and Great American also filed AI-specific exclusions or limitations during this period.

The 80% approval rate reflects regulatory comfort with the exclusion framework, not regulatory indifference. State DOI reviewers have largely accepted the argument that AI represents a novel and unquantified risk that existing policy language was never designed to price. From a regulatory perspective, allowing carriers to exclude a risk they cannot price is preferable to forcing them to absorb losses that could threaten solvency.

Carrier Form / Approach Lines Affected Scope
ISO / Verisk CG 40 47 01 26 CGL (Coverage A + B) Total generative AI exclusion
ISO / Verisk CG 40 48 01 26 CGL (Coverage B only) Personal & advertising injury only
W.R. Berkley PC 51380 D&O, E&O, Fiduciary Absolute AI exclusion
Hamilton HAM-AI-2025 E&O, Cyber Sublimit (25-50% of policy limit)
Berkshire / Chubb / Travelers ISO-based filings CGL State-by-state AI exclusion filings

Greenberg on Mythos: The CEO Who Named the Trigger

Chubb’s Q1 2026 earnings call produced the most candid executive commentary on AI risk from any carrier CEO. Greenberg did not speak in generalities. He named Anthropic’s Claude Mythos model specifically, framing it as a catalyst for a new category of cyber-adjacent exposure.

“The arms race is on,” Greenberg told analysts. He explained that Mythos has “lowered the threshold for vulnerability,” allowing minor security gaps to be “aggregated in a much more insightful way.” Tools like Mythos can read and analyze code at scale, meaning they can “find vulnerabilities maybe even before suppliers do.” The catch, as Greenberg noted: “It doesn’t mean the patch has been created.”

He identified middle-market organizations as the primary vulnerability, calling them the “biggest meatball” for attackers because they possess more financial resources than small firms but demonstrate weaker cybersecurity practices and “weaker perimeters” than large enterprises. This is precisely the market segment where Chubb holds significant commercial lines market share.

Greenberg warned that while cyberattacks currently involve human operators (“humans are in the cockpit when they’re using agentics so far”), fully autonomous offensive AI capabilities are “just around the corner.” The implication for underwriting: the current loss experience for cyber and AI-adjacent lines is already stale. The risk profile of the insured base is changing faster than the data in the loss triangles.

This is the same CEO whose company is simultaneously filing to exclude AI-related damages from its own commercial policies. When the head of the world’s largest publicly traded P&C insurer says the arms race is on, the exclusion filings stop looking like routine product adjustments and start looking like strategic risk shedding at scale.

The Litigation Trajectory That Drove the Exclusion Wave

Carrier exclusion strategies did not emerge in a vacuum. The filings followed a sharp acceleration in AI-related litigation that moved from theoretical to material in under two years.

Year-over-year AI-related legal filings accelerated to approximately 137% in 2024-2025, roughly doubling the 59% growth rate observed in 2023-2024. A Gallagher Re/MIT report titled “Smart Systems, Blind Spots: Rethinking Insurance for the AI Era” documented a 978% increase in generative AI-related lawsuits in the United States between 2020 and 2025, with cumulative filings exceeding 700 cases. Skadden reported that AI-related securities class actions have outpaced other categories, with plaintiffs targeting companies for misrepresentations about AI capabilities, overstated efficiency gains, rebranding legacy technology as AI, and concealing licensing or performance issues.

Gartner projected that more than 2,000 legal claims linked to “death by AI” incidents will be brought worldwide by the end of 2026. Whether that projection proves accurate or not, the trajectory gave carrier actuaries and product managers the data they needed to justify exclusion filings to their boards and regulators.

The litigation breaks into four categories that map directly to policy coverage sections:

  • Employment discrimination via AI systems: Claims alleging that employer-deployed AI in hiring, performance evaluation, or termination decisions produced discriminatory outcomes. These fall under CGL Coverage A and employment practices liability.
  • Intellectual property violations: AI-generated content that infringes copyrights, trademarks, or trade secrets. These trigger Coverage B personal and advertising injury provisions.
  • Product liability for AI outputs: Physical harm or property damage caused by autonomous systems, robotics, or AI-generated instructions. These implicate both CGL Coverage A and products/completed operations.
  • “AI washing” securities claims: D&O exposure from companies overstating their AI capabilities, understating risks, or misrepresenting the role of AI in their operations. These hit the management liability lines where Berkley’s PC 51380 applies.

Each category is growing. No category is shrinking. Actuaries reviewing loss development on these claim types cannot identify a scenario where frequency decelerates in the near term.

The Silent Coverage Gap: What Enterprise Risk Managers Are Missing

The most consequential effect of the exclusion wave is not what it removes from policies. It is what policyholders do not yet know they have lost. Brokerages Aon, Gallagher, and Lockton have all publicly flagged the implications: companies deploying AI agents could discover that “previously assumed coverage is simply gone.”

Most commercial policyholders have not reviewed their CGL policies for AI-specific endorsements. Many renewal packages in Q1 and Q2 2026 now include CG 40 47 or CG 40 48 attachments that most risk managers have not read closely, let alone modeled for coverage impact. The exclusions are buried in endorsement schedules, not flagged in coverage summaries.

This creates what the industry has termed a “silent AI” coverage gap, paralleling the “silent cyber” dynamic that reshaped the cyber insurance market between 2017 and 2021. In that earlier cycle, carriers realized their CGL and property books contained unpriced cyber exposure, filed exclusions to remove it, and a standalone cyber market emerged to fill the gap. The AI exclusion cycle is following the same playbook, but at a compressed timeline.

The exposure categories most likely to catch enterprises by surprise include:

  • Marketing and content teams using AI-generated copy that reproduces copyrighted material. No Coverage B protection under CG 40 48.
  • HR departments deploying AI screening tools that produce disparate impact outcomes. Potential gaps in both CGL and EPL coverage depending on endorsement configuration.
  • Product development teams embedding generative AI in customer-facing products. No coverage under CG 40 47 for downstream harm.
  • Board-level AI governance failures exposed by shareholder suits. No D&O protection under Berkley’s PC 51380.
  • Professional services firms relying on AI-generated analysis or recommendations. No E&O coverage under absolute exclusion forms.

For actuaries consulting on enterprise risk management programs, the audit priority is straightforward: pull every commercial policy renewal from the past six months, search for AI-related endorsements, and map the coverage that existed twelve months ago against what exists today. The delta is the unpriced exposure the enterprise now retains.

The Standalone Market Response

Where standard carriers retreat, specialty markets advance. A new class of standalone AI liability products is emerging to fill the coverage gap, backed by a mix of startups, specialty carriers, and reinsurer-supported capacity.

Corgi reached a $1.3 billion valuation with AI liability coverage targeting model-specific failure events, hallucination liability, and regulatory defense costs. Armilla launched in 2025 with coverage requiring ongoing model quality assessments. Testudo launched in January 2026 with a claims-made product targeting mid-to-large enterprises deploying generative AI. Mayflower Specialty and Embroker have also entered the market with tailored AI coverage offerings. Munich Re has offered AI-specific insurance products since 2018, though primarily targeting AI startups rather than enterprise adopters.

Coverage limits in the standalone market range from $2 million to $50 million, with premiums spanning “a few hundred dollars to several hundred thousand dollars annually” depending on deployment scope, model governance maturity, and claims history. The Deloitte Center for Financial Services projects that by 2032, insurers could write approximately $4.7 billion in annual global AI insurance premiums, at a compound annual growth rate of roughly 80%.

The structural challenge for this market is the same one that hampered early standalone cyber: pricing without credible loss data. With fewer than 700 cumulative AI-related lawsuits through 2025 and no mature loss triangles, actuaries building rate plans for standalone AI products are relying on scenario-based modeling, expert elicitation, and analogies to the cyber market development arc rather than actuarial experience rating. This is inherently fragile. The first large AI liability verdict or settlement will reprice the entire standalone market overnight, just as the 2017 WannaCry and NotPetya events forced a wholesale recalibration of cyber pricing.

The Actuarial Pricing Signal

From an actuarial perspective, the deploy-and-exclude paradox contains a clear pricing signal: the carriers closest to the risk, the ones building and deploying AI systems at scale, have concluded that they cannot price the external exposure. This is not a coverage gap born of ignorance. It is a deliberate judgment by the industry’s most sophisticated risk quantifiers that the distribution of AI-related losses is too uncertain, too fat-tailed, and too correlated across lines to retain on standard commercial books.

Several actuarial implications follow:

Loss development factors are unreliable. Traditional LDFs require a stable development pattern. AI-related claims have no established pattern. The lag between AI deployment and claim emergence could be months (content copyright infringement) or years (long-tail product liability from AI-generated medical advice). Without stable development, IBNR estimates for AI exposure are speculative at best.

Correlation across lines creates aggregation risk. A single AI system failure can trigger CGL, D&O, E&O, cyber, and employment practices claims simultaneously. Carriers pricing these lines independently are underestimating the correlation. The exclusion wave partly reflects this recognition: rather than trying to model the correlation, carriers are removing the exposure from multiple lines at once.

The cyber precedent is instructive but imperfect. Standalone cyber took roughly a decade from the first exclusion filings (mid-2010s) to reach pricing stability (mid-2020s), and even then only after multiple loss events recalibrated models. AI liability may compress that timeline because the litigation trajectory is steeper, but the underlying uncertainty remains comparable. Actuaries building AI liability rate plans should expect 3-5 years of elevated loss ratio volatility before credible experience data emerges.

Internal AI deployment creates a moral hazard question. Carriers using AI to price, underwrite, and adjudicate claims while excluding AI risk from the policies they sell face an asymmetric information problem. If a carrier’s own AI system produces a flawed underwriting decision that harms a policyholder, which policy responds? The E&O policy? The CGL policy? Both are now potentially excluded. This creates uncharted territory for coverage litigation and actuarial reserving.

Why This Matters

The deploy-and-exclude paradox is not a temporary market dislocation. It is the beginning of a multi-year structural shift in how the insurance industry handles technology risk. Three dynamics will shape the next 12 to 24 months.

First, the exclusion wave will continue to expand. Carriers that have not yet filed AI-specific endorsements are watching the 80% approval rate and the absence of regulatory pushback. By year-end 2026, we expect the majority of top-25 P&C carriers will have filed some form of AI exclusion across at least one commercial line. The holdouts will be carriers that view AI coverage as a competitive differentiator, not because they have solved the pricing problem.

Second, the standalone AI liability market will grow but remain fragile. The $4.7 billion Deloitte projection assumes a smooth growth curve. The reality will be lumpy, driven by individual large losses that either validate or destroy early pricing assumptions. The first AI-related verdict exceeding $100 million will be the market’s defining event, analogous to the role that major data breaches played in shaping cyber insurance pricing.

Third, the regulatory landscape is still forming. The Colorado AI Act bias audit requirements taking effect July 1, 2026 will force carriers to document their own AI governance practices. The state AI law patchwork is creating compliance complexity that will factor into both internal deployment costs and external exclusion strategies. If state regulators begin requiring affirmative AI coverage in certain lines, as some consumer advocacy groups are now urging, the exclusion strategy will face its first serious constraint.

For actuaries, the action items are concrete. On the consulting side: audit every commercial client’s renewal for AI-specific endorsements and quantify the retained exposure. On the carrier side: model the correlation between internal AI deployment risk and external AI exclusion strategy, because regulatory and litigation scrutiny of that contradiction is inevitable. On the reserving side: begin building AI-specific claim taxonomies now, even without credible loss data, so that when experience does emerge, the reserving infrastructure is ready to process it.

The carriers that deploy AI most aggressively are the same ones shedding AI risk most aggressively. That is not a coincidence. It is a signal that the industry’s best-informed participants believe AI risk is real, material, and currently uninsurable at standard commercial rates. Whether that view proves correct or overly conservative will determine the shape of the P&C market for the next decade.

Further Reading

Sources