Standard P&C carriers are aggressively deleting "silent" AI coverage from commercial renewals, forcing an unpriced liability crisis directly onto the balance sheets of enterprise deployers. This intelligence brief breaks down the exact ISO exclusion forms driving this retreat and provides the proprietary competitor intelligence you need to monetize the $4.7 billion specialty market stepping in to fill the void.

Below the paywall is our full Intelligence Brief on this subject: a dense tactical playbook designed for strategy, product, and go-to-market executives at carriers and brokerages.

Looking for the free summary version before you upgrade? Click here.

This intelligence brief isn’t built on recycled news; it is the synthesis of a massive, multi-dimensional research effort. To uncover the true scale of the “Silent AI” coverage gap, we aggregated and analyzed 190 distinct industry sources, cross-referencing public data with highly restricted, proprietary insights. The depth and diversity of the intelligence informing this brief are unmatched, drawing directly from:

By layering these diverse sources, from the macro-economic signals of global reinsurance renewals to the micro-level realities of specific ISO exclusion forms, we have built an intelligence brief that eliminates the noise. The full brief provides the exact policy forms being used to drop coverage, a breakdown of the legacy carriers driving the trend, and the proprietary broker playbook for navigating the only three standalone AI liability markets currently writing this risk.

The dangerous assumption that broad Commercial General Liability (CGL) and management liability policies will act as a safety net for artificial intelligence risks officially died in early 2026. As a new liability class forms around enterprise AI deployment, the insurance industry is aggressively and systematically unbundling this exposure from standard commercial lines. This coverage gap became structural as legacy carriers began stripping coverage from their books rather than absorbing the unpriced risks associated with algorithmic decision-making, leaving enterprises highly exposed.

The property and casualty industry is moving swiftly to eliminate “silent AI” exposure before catastrophic losses hit, learning directly from the painful reserve distortions of the “silent cyber” crisis. Swiss Re has explicitly warned of this emerging “silent AI” exposure, likening it to the silent cyber problem where insurers unknowingly assume massive liabilities under existing policies that were never designed for algorithmic failures. Rather than waiting for courts to interpret ambiguous coverage, conventional insurers are actively withdrawing from silent AI exposure while a small cluster of specialist Excess & Surplus (E&S) underwriters moves into the gap.

As of January 1, 2026, new Verisk/ISO exclusion endorsements have given carriers a standardized, low-friction mechanism to carve generative AI-driven losses entirely out of CGL policies. Prior to these endorsements, carriers faced the operational hurdle of negotiating bespoke manuscript exclusions. Now, ISO’s GenAI endorsements are deliberately modular, allowing carriers to tune their risk appetite and explicitly exclude the specific liability towers where AI friction is most acute without relying on future ambiguity.

The core ISO filings explicitly remove coverage across multiple liability towers, providing carriers with precise tools to neutralize the most litigable AI risks. Form CG 40 47 excludes bodily injury, property damage, and personal and advertising injury arising from generative AI across both occurrence and claims-made CGL coverage. Meanwhile, Form CG 40 48 acts as a surgical tool removing only Coverage B, which is the immediate pressure point for highly litigable claims. The necessity of this Coverage B carve-out is underscored by the explosion in GenAI litigation; patent infringement now accounts for 11.9% of cases, copyright infringement represents 11.2%, and personal injury tied to privacy violations makes up 10.2%.

Furthermore, ISO Form CG 35 08 serves as a forward warning signal by stripping products and completed operations coverage, shifting massive accumulation risk back to the enterprise when AI is embedded as a feature. Once generative AI is embedded in the products or completed work that an insured delivers to its own customers, the liability shifts from a simple “bad statement” to downstream harm theories. Insurers recognize they cannot safely price “AI as a shared failure mode” inside a baseline general liability policy, particularly when multiple unrelated insureds depend on the same upstream foundation model’s behavior.

The deployment of these exclusions is not a fringe movement; it is being driven by the largest market-makers in the P&C industry, with state regulators already approving more than 80 percent of these filings. Berkshire Hathaway, Chubb, and Travelers began submitting applications to drop this coverage last fall, and these provisions are already in effect. The regulatory environment has been highly accommodating to this carrier retreat, with the highest volumes of approvals passing through in Florida, Connecticut, and Maryland. A growing list of carriers, including AIG, Great American, Hamilton Insurance Group, and Philadelphia Indemnity, have subsequently filed their own AI exclusions with state regulators.

This aggressive exclusion wave creates a glaring competitive paradox, as carriers like Chubb are publicly touting their internal AI transformations while simultaneously ensuring their commercial clients possess no coverage for deploying those exact same technologies. In a recent shareholder letter, Chubb CEO Evan Greenberg committed to automating 85% of key underwriting and claims functions by 2028, a move that involves cutting approximately 8,600 jobs—roughly 20% of its global workforce—to improve combined ratios by 1.5 points. Despite leading the industry in AI adoption to drive its own record Q1 2026 net income of $2.32 billion and a stellar 84.0% combined ratio, Chubb is actively utilizing the new ISO endorsements to shed the exact AI liability exposure its commercial clients are now facing.

Some carriers are pushing beyond targeted CGL carve-outs to deploy absolute AI exclusions that act as a complete “kill switch” across professional and management lines. WR Berkley has introduced Form PC 51380, an absolute AI exclusion designed to eliminate coverage across Directors & Officers (D&O), Errors & Omissions (E&O), and fiduciary liability for any claim arising from the use, deployment, or development of AI by any person or entity. The breadth of this language does not just target bad algorithmic outputs; it explicitly targets AI-generated content, the failure to detect third-party AI content, “inadequate” corporate AI policies, and regulatory violations. This unforgiving posture functions as a practical withdrawal from the class of insureds that utilize AI.

Ultimately, standard admitted carriers are using these new exclusions to refuse to subsidize AI ambiguity, effectively forcing the exposure out of the admitted market and into specialty channels. By deleting this coverage at renewal, legacy carriers are stranding commercial deployers and creating the exact conditions required for a standalone specialty market to form. Carriers and MGAs that build standalone AI liability underwriting capacity now, during this early price discovery window, will establish structural positions in a market that Deloitte projects will reach $4.7 billion in annual premiums by 2032.

The aggressive withdrawal of standard admitted carriers from artificial intelligence exposures exposes a dangerous reality for commercial insureds: the specialty insurance safety net is completely empty. As standard liability policies such as Commercial General Liability (CGL) systematically drop coverage for algorithmic failures, enterprise risk managers naturally assume their existing specialty lines will fill the void. However, recent data proves that AI liabilities currently sit in the uninsurable blind spots between every major existing property and casualty product, forcing an unpriced liability crisis directly onto the balance sheets of the enterprises deploying the technology.

The structural driver forcing this liability crisis onto enterprise balance sheets is a highly restrictive vendor contract design that completely shields foundational model developers from downstream harm. Commercial insureds operating under the assumption that the creators of AI models will bear the responsibility for their failures are operating under a false sense of security. All four major foundational AI vendors—OpenAI, Anthropic, Google, and Microsoft—strictly cap their own liability at 12 months of fees paid and expressly disclaim all consequential damages in their standard terms. Because no major vendor offers performance warranties, the enterprise customer—the deployer—bears the residual legal liability for virtually everything the AI system produces.

This transfer of legal responsibility is colliding with a massive explosion in litigation, as generative AI-related lawsuits in the United States surged by 978 percent between 2012 and 2025. According to a March 2026 white paper published by Gallagher Re in association with the Massachusetts Institute of Technology (MIT) and Testudo Global, the litigation trajectory is accelerating rapidly. The year-over-year growth in lawsuit filings jumped from 59 percent between 2023 and 2024 to a staggering 137 percent between 2024 and 2025. Cumulative GenAI-related lawsuits in the U.S. have now climbed past the 700 mark just between the years 2020 and 2025, proving this exposure is actively bleeding into the legal system.

The specific claims driving this litigation wave demonstrate that AI exposure is a content-at-scale risk that severely impacts intellectual property and privacy rights. The Gallagher Re and Testudo data reveals the leading claim categories are patent infringement (11.9 percent of cases), copyright infringement (11.2 percent), and personal injury tied primarily to privacy violations and data misuse (10.2 percent). The financial severity of these claims was benchmarked in August 2025 when Anthropic settled copyright claims at approximately $3,000 per infringing work, establishing a terrifying damages precedent for enterprises deploying generative AI to produce thousands of marketing or communication assets. Consequently, a recent Gallagher survey found that 57 percent of companies now identify AI errors, misinformation, and hallucinations as their leading overall risk concern.

Executives mistakenly relying on Technology Errors & Omissions (Tech E&O) policies are discovering that these instruments protect the software creators, not the downstream enterprise users. Technology E&O was purposefully designed to protect technology providers and developers. This fundamental structural limitation means that the vast majority of non-tech enterprises—such as retailers, logistics firms, or financial services companies—that utilize third-party AI tools are left entirely uncovered for financial losses, defamation, or unauthorized data disclosures caused by those tools’ outputs.

Similarly, traditional Cyber insurance structurally fails to cover generative AI failures because it treats AI solely as an attack vector rather than a direct source of liability. While a cyber insurance policy is designed to respond to an AI-enabled external threat, such as an attacker using AI-assisted phishing to breach a network, it does not cover liabilities arising from a company’s own AI outputs. If an enterprise’s AI chatbot hallucinates, generates defamatory content, infringes on intellectual property, or is manipulated via “malicious prompts” into issuing unauthorized mass refunds, the cyber policy will not respond.

The escalation of autonomous AI offensive capabilities further degrades existing cyber defenses, turning enterprise AI agents into highly vulnerable attack surfaces themselves. In April 2026, Anthropic’s Claude Mythos Preview model demonstrated the ability to autonomously discover thousands of high-severity zero-day vulnerabilities across major operating systems, successfully converting 72.4 percent of them into working exploits. Threat actors are simultaneously weaponizing these capabilities; Aon’s AI Risk 2026 report found that AI-generated phishing achieves click-through rates of approximately 54 percent, compared to just 12 percent for traditional attacks. As AI agents gain access to financial and customer service workflows, prompt injection and behavioral manipulation could trigger real-world loss events that blur the line between operational error and cybercrime.

The culmination of these coverage gaps and accelerating threats leads to the reinsurance sector’s ultimate fear: systemic, cross-border accumulation risk. Unlike traditional natural catastrophe scenarios—such as hurricanes or wildfires—that are constrained by geographic or sectoral boundaries, AI failures propagate instantly across all borders. A flaw, degradation, or data poisoning event in a single, widely adopted foundational model embedded in thousands of enterprise workflows could trigger simultaneous, highly correlated claims across entirely unrelated industries and policyholders.

Because the global insurance market cannot absorb thousands of highly correlated claims from a single foundational model failure, leading advisors are demanding structural changes to how AI risk is classified. Aon’s Kevin Kalinich framed the arithmetic plainly: while the global insurance industry can easily absorb a $400 million to $500 million loss stemming from one individual company’s AI deployment failure, it absolutely cannot handle 1,000 or 10,000 correlated losses stemming from a single upstream model failure. Given this reality, leading advisors like Lockton Re are now urgently calling for the industry to treat AI as its own separate peril and distinct risk class. Until that foundational shift enables proper actuarial reserving and reinsurance treaty development, standard carriers will continue to exclude the risk, leaving the enterprise deployer holding the liability.

The aggressive retreat of standard admitted carriers from the generative AI space has catalyzed the formation of a highly specialized Excess & Surplus (E&S) market. Deloitte projects the global AI insurance premium market could reach $4.7 billion annually by 2032, triggering a massive land grab for orphaned deployer liability. Because standard actuarial loss history does not yet exist for algorithmic failures, conventional P&C underwriting desks lack the technical expertise to assess model architectures, hallucination risk profiles, and training data provenance. Consequently, this multi-billion-dollar gap is being filled by specialized Managing General Agents (MGAs) and dedicated reinsurer units. Currently, there are exactly three standalone AI liability insurance products in the market that are actively bindable to fill this gap.

Armilla AI has established itself as the market pioneer by operating as the first Lloyd’s Coverholder dedicated exclusively to underwriting AI liability. Having initially launched its AI Liability Insurance product in April 2025 with Chaucer Group as the lead underwriter, Armilla expanded its offering in January 2026 to provide robust $25 million limits per insured under a broad all-risks framework. The facility aggregates capacity from a syndicate of heavyweights, including AXIS Capital, Swiss Re, Convex, and Greenlight Re. This surplus lines product covers third-party claims arising from AI hallucinations, model drift, inaccurate outputs, data leakage, and algorithmic bias, while also covering defense costs and insurable fines tied to regulatory violations under the EU AI Act and Colorado AI Act.

The defining characteristic of Armilla’s approach is its strict “double trigger” underwriting requirement, which mandates an independent technical certification of the insured’s AI systems prior to binding. To successfully invoke coverage, an AI system’s underperformance must first occur and must then lead to a measurable financial loss or a legal claim. Because this involves an intensive, judgment-heavy technical audit—informed by more than 500 evaluations Armilla has already conducted across regulated industries—the product is uniquely tailored for AI scale-ups and Fortune 1000 enterprises that possess mature, heavily documented AI architectures. CEO Karthik Ramakrishnan has explicitly noted that most insurance policies were not designed for generative AI or AI agents, making this rigorous assessment model a necessity for companies embedding these systems at scale in sectors like financial services, healthcare, and human resources.

Taking a radically different, lower-friction approach to risk selection, Testudo Global bypasses the technical audit bottleneck entirely by pricing risk through a proprietary database of global AI litigation. Operating as a Lloyd’s Lab Cohort 14 participant, Testudo went live in January 2026 with a claims-made E&S policy aimed specifically at U.S. enterprises deploying generative AI. Supported by A+ Superior rated Lloyd’s capacity from the Apollo ibott 1971, Atrium, and QBE syndicates, Testudo offers limits of up to $9.25 million per insured. Instead of evaluating technical model performance metrics via invasive IT integration, Testudo utilizes its proprietary AI Risk Engine to generate risk scores and bespoke risk summaries based on real-world claim patterns extracted from historical AI lawsuits.

By explicitly refusing to underwrite AI developers or vendors, Testudo has perfectly positioned its product as the fastest replacement path for middle-market commercial deployers losing their legacy coverage. The strategy here is purely focused on downstream enterprise liability, positioned as a direct response to the ISO CGL generative AI exclusions. Because brokers and enterprises can register interest directly through Testudo’s website without opening their systems to an external audit, the submission process is remarkably fast and scalable. As Testudo CEO George Lewin-Smith articulated, while many carriers and syndicates deem the market impossible due to a lack of traditional data, Testudo relies on its litigation data advantage to confidently price the exposure. This makes it an ideal landing spot for wholesale brokers scrambling to replace coverage after their clients receive an AI exclusion endorsement on their renewals.

Munich Re possesses the longest operating history in this niche, utilizing a parametric-like framework that relies on highly measurable, continuous performance data rather than lengthy post-incident investigations. Launching its aiSure product originally in 2018 as a performance warranty for AI model underperformance, Munich Re provides direct AA-rated capacity rather than operating through an MGA structure. The policy uniquely pays out financial losses—including lost revenue, business interruption, and legal damages—when an AI system’s error rate exceeds pre-established key performance indicator (KPI) baselines. Like Armilla, Munich Re mandates thorough technical due diligence before coverage can be bound, restricting access to clients with mature, measurable AI deployments where performance metrics are already established and actively monitored.

In February 2026, Munich Re significantly scaled its product’s market reach by partnering with Mosaic Insurance to distribute up to 15 million (EUR/USD/CAD) in coverage globally. This partnership leverages Mosaic’s global cyber specialist network to underwrite and market the product directly to AI developers and vendors worldwide, a stark contrast to Testudo’s deployer-only focus. Dennis Bertram, Head of AI Underwriting at Mosaic, clarified that this coverage is entirely focused on whether the AI model’s outputs are actually accurate, rather than focusing on system uptime or traditional cyber incidents. This creates a dedicated safety net that evaluates the model itself, what it does, and how its outputs are used, rather than focusing purely on the insured’s respective industry.

While a fourth standalone product from Vouch exists at a smaller scale for AI startups, the broader enterprise casualty market will be defined by the capacity generated by Armilla, Testudo, and Munich Re. Vouch launched its AI Insurance in January 2024 to cover AI errors and omissions, algorithmic bias, intellectual property infringement, and regulatory investigation defense costs. However, the product is targeted exclusively at early-stage AI startups and operates outside the typical enterprise casualty placement that commercial brokers handle. Ultimately, as E&S wholesaler executives like Andrew Kelly project, a massive dedicated insurance sector with its own MGAs, claims professionals, and policy forms is actively forming over the next five to ten years to address this gap. The specialized underwriting judgment being deployed by these select market-makers proves that AI is simultaneously the tool carriers are using to automate their own operations and the multi-billion-dollar liability class they are terrified to underwrite directly.

The migration of AI liability into the Excess & Surplus (E&S) market is no longer a future-state theory; it is an immediate execution challenge that requires a highly specific, data-driven playbook. With over 80 percent of state regulators approving Verisk/ISO AI exclusions for standard commercial policies, the coverage gap is live and the traditional broker channel is largely unprepared. However, this disruption creates a massive structural opportunity for professionals who know how to navigate the emerging specialty landscape. To capture their share of a market that Deloitte projects will reach $4.7 billion in annual premiums by 2032, brokers and carriers must abandon legacy assumptions and execute targeted placement and underwriting strategies.

For wholesale placement professionals and retail brokers, the immediate priority is securing appointments with the three standalone AI liability markets to establish structural positions in this specialty land grab. Just as the brokers who built early relationships with cyber MGAs in 2012 and 2013 captured outsized market share, the window is now open to dominate AI liability placement during its critical price-discovery phase. Because the three active markets—Testudo Global, Armilla AI, and Munich Re—use fundamentally different underwriting methodologies, a broker’s primary job is precise client-matching. Applying the wrong submission strategy or failing to understand the required technical auditing thresholds will result in immediate declinations.

When a middle-market or large U.S. enterprise account receives an ISO AI exclusion endorsement on their CGL renewal, Testudo Global provides the fastest, most scalable path to replacement coverage. Backed by A+ Superior rated Lloyd’s capacity from the Apollo ibott 1971, Atrium, and QBE syndicates, Testudo offers claims-made E&S limits up to $9.25 million per insured. Crucially for brokers, Testudo explicitly targets enterprise deployers and requires absolutely no pre-deployment technical audit or invasive integration with the insured’s AI systems. Instead, the company utilizes a proprietary AI Risk Engine to generate risk scores based on a global database of AI litigation, allowing brokers to secure coverage for clients without forcing them to open their proprietary IT infrastructure to external evaluation.

For Fortune 1000 enterprises and AI scale-ups that possess mature, heavily documented AI architectures, Armilla AI offers necessary capacity through a rigorous technical auditing model. As the first Lloyd’s Coverholder dedicated exclusively to AI liability, Armilla provides $25 million limits per insured under a broad all-risks framework, drawing capacity from Chaucer Group, AXIS Capital, Swiss Re, Convex, and Greenlight Re. The facility covers third-party claims arising from hallucinations, model drift, algorithmic bias, and regulatory violations under the EU AI Act and Colorado AI Act. However, Armilla operates on a strict “double trigger,” meaning an independent technical certification of the client’s AI systems is required before binding. Brokers must prepare these clients for a deep technical evaluation, making this product highly suited for regulated sectors like financial services, healthcare, and human resources where performance characteristics are strictly measured.

Brokers representing AI developers, vendors, and highly sophisticated deployers should target the Munich Re and Mosaic Insurance facility, which utilizes a parametric-like framework based on performance warranties. Having launched its aiSure product in 2018, Munich Re offers up to 15 million (EUR/USD/CAD) in direct AA-rated capacity globally. This policy pays out financial losses—including lost revenue, business interruption, and legal damages—when an AI system’s error rate exceeds pre-established KPI baselines, enabling rapid claims settlement. As Dennis Bertram, Head of AI Underwriting at Mosaic, clarified, this coverage focuses purely on the accuracy of the AI model itself and how its outputs are used, requiring thorough technical due diligence before coverage can be bound.

For standard and specialty carriers considering writing controlled, affirmative AI coverage, underwriting evaluations must fundamentally shift from asking “Do you use AI?” to demanding strict, auditable operational controls. Carriers cannot safely underwrite AI exposure without requiring insureds to implement a documented AI governance framework. Underwriters must evaluate vendor and last-mile controls, demand role-based restrictions on which tools can be used, and verify that controls exist to block the upload of sensitive data into third-party models. Furthermore, carriers must assess “agentic oversight,” which includes monitoring autonomous agents for runtime threats, establishing performance monitoring for algorithmic drift, and maintaining clear boundaries between assisted human workflows and fully automated decisions.

Carriers must also enforce a “claims-proof evidence” standard in their underwriting files, moving beyond simple attestations to mandate structured telemetry that can reconstruct an AI failure. In practice, AI disputes will turn on whether investigators can retrace the digital sequence of events; without retention logs, AI claims become litigation-shaped and inconsistency-prone. The minimum evidence stack required should include identity and device posture, access logs, data classification context, prompt text, output metadata, and the specific model version in use at the time of the incident. The legal precedent for strictly enforcing these controls was cemented in Travelers Property Casualty Company of America v. International Control Services Inc., where a federal court voided a cyber policy from inception because the insured misrepresented its deployment of multi-factor authentication (MFA). If an insured misrepresents its AI telemetry or governance controls, carriers must have the contractual leverage to rescind coverage.

Furthermore, mounting regulatory pressure from state insurance departments requires that any AI model used by carriers or their insureds must be transparent, explainable, and free of algorithmic bias. The NAIC has adopted a Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, which establishes templates for AI governance, risk management controls, and third-party system oversight. Simultaneously, states like Colorado now require insurers to submit detailed compliance reports proving their frameworks prevent unfair discrimination and statistical bias. Black-box AI has officially become a compliance liability; underwriting teams must be able to see and explain their decision logic, or they will face growing regulatory exposure and enforcement actions.

While brokers secure placements and carriers tighten underwriting guardrails, the entire market is operating under the shadow of a massive, unresolved systemic threat: cross-border accumulation risk. Unlike traditional natural catastrophes—such as hurricanes or wildfires—that are constrained by geographic or sectoral boundaries, AI failures propagate instantly across all borders. Reinsurance executives are deeply concerned about a scenario where a flaw, degradation, or data poisoning event in a single, widely adopted foundational model (such as an OpenAI, Google, or Anthropic system) triggers simultaneous, highly correlated claims across thousands of unrelated enterprise workflows.

Because the global insurance market cannot absorb thousands of highly correlated claims from a single foundational model failure, leading advisors are demanding a structural change to treat AI as its own separate peril. Aon’s Kevin Kalinich framed the arithmetic plainly: the global insurance industry can easily absorb a $400 million to $500 million loss stemming from one individual company’s AI deployment failure, but it absolutely cannot handle 1,000 or 10,000 correlated losses stemming from a single upstream model failure. Given this reality, leading advisors like Lockton Re and Gallagher Re are urgently calling for the industry to classify AI as a distinct risk class. Until this foundational shift enables proper actuarial reserving and reinsurance treaty development, standard carriers will continue their aggressive retreat, leaving agile E&S markets to dictate the pricing of this multi-billion-dollar exposure.

The AI exclusion wave is a structural repricing of who bears liability for algorithmic failures, and the answer is increasingly the enterprise deployer. Standard admitted markets have made their position clear through ISO endorsements, absolute exclusions, and accelerating state-level approvals. The specialty market forming to replace that coverage is real but nascent. Organizations looking to move beyond this brief and into more specific work, whether on competitive positioning, coverage gap analysis, or market entry, may reach out to our partner Emerging Strategy directly for customized strategic advisory services.

P&C Insurance Executive Intelligence is for strategy, product, and executive leaders in carriers, reinsurers, and insurance platforms navigating commercial lines disruption.

Ping us at [email protected] if you’d like to learn more, explore Institutional Subscriptions, or would like to partner in other ways.

The Intelligence Council is a next-gen B2B media and business intelligence platform built for people who make strategy, allocate capital, and carry operating risk.