The insurance sector intersects with AI in two directions. As a deployer, insurers apply AI to claims automation, underwriting, and fraud detection. As underwriters of risk, insurers have begun to shape a market for "insuring AI" — products that cover, or specifically exclude, losses caused by AI systems. Through 2025 and into 2026, established carriers moved to exclude AI-driven losses from standard commercial policies while a set of specialist carriers entered to provide dedicated AI-liability cover.
Adoption patterns
Insurance pricing is intrinsically algorithmic, and extending AI-driven pricing raises Algorithmic Pricing and Antitrust questions specific to regulated insurance markets. The National Association of Insurance Commissioners (NAIC) and state insurance regulators have begun issuing guidance on AI use in underwriting.
Insuring AI
A market for products covering losses caused by AI systems began forming in 2025. On May 11, 2025, reporting described insurers launching cover for losses caused by AI chatbot errors. Armilla, a Y Combinator–backed firm, offered AI-system risk coverage positioned as an "AI liability insurance" product. Demand for such coverage tracks the spread of enterprise AI deployment and exposure to AI errors, hallucinations, and discriminatory outputs.
AI-specific exclusions and caps
Through the April 2026 commercial-insurance cycle, established carriers moved to carve AI-driven losses out of standard policies. Berkshire Hathaway, Chubb, Travelers, AIG, Tokio Marine, W.R. Berkley, and Fairfax Financial all filed AI-specific exclusion language on commercial general-liability forms (Source: insurancejournal.com).
Per Wolfe Research's analysis of regulatory filings (Apr 23, 2026), state regulators approved more than 80% of insurer requests for AI exclusions from corporate policies. Florida, Connecticut, and Maryland approved the highest counts. California, New York, and Texas — home to many AI providers — had not yet approved the ISO templates, so traditional policies in those states still covered AI losses at the time of reporting (Source: theinformation.com). The Insurance Services Office (ISO), the private body that sets industry standards, published AI-exclusion endorsements in the second half of 2025, after which insurers "meaningfully accelerated" filings, per Wolfe Research's Tracy Benguigui. Wolfe Research attributed insurer caution partly to litigation pressure, counting roughly 800 consumer lawsuits against businesses using AI products in 2025, up 140% year over year.
Caps appeared on adjacent lines as well. QBE and Beazley placed AI-driven cyber caps on standalone cyber policies (Source: reuters.com). Per Aon's Kevin Kalinich and Lockton's Preet Gill, insurers were expected to extend exclusions to cybersecurity policies and errors-and-omissions policies, so the carve-out was unlikely to remain limited to general liability. The Financial Times separately reported insurers moving to cap AI-related losses in cybersecurity policies (Source: ft.com).
The exclusion wave has been described as patterned on the earlier treatment of cyber risk: traditional policies covered cyber losses by silence, and insurers later argued they should not have, so carriers moved to exclude AI losses explicitly rather than by silence. Codestrap CEO Connor Deeks stated: "Companies are being pushed to use AI at unsustainable rates. The risks will become unsustainable and translate to financial losses."
With AI-specific risks formerly covered under general commercial liability now carved out, buyers must seek bespoke AI cover from specialist carriers (Armilla, Artificial Intelligence Underwriting Company, Corgi, Mayflower Specialty, Embroker, Munich Re). The exclusions price in regulatory uncertainty: carriers face difficulty underwriting AI-driven claims at scale until state and EU AI-discrimination liability regimes are clearer. Commentary has framed insurer non-coverage as a governance lever that may discipline enterprise AI deployment in the absence of formal regulation.
Specialist AI-liability carriers
A set of carriers entered specifically to underwrite AI risk that traditional insurers were carving out.
| Carrier | HQ / structure | Coverage limit | Notable clients / details |
|---|---|---|---|
| Artificial Intelligence Underwriting Company | San Francisco | up to $50M | ElevenLabs |
| Armilla | Canada; Lloyd's of London is the primary risk-bearer | up to $25M | Dozens of clients |
| Munich Re | Germany; world's largest reinsurer | n/a | First major reinsurer with standalone AI policies |
| Corgi | Startup; tech-startup focus | up to $2M | Annual premiums $hundreds–$hundreds-of-thousands |
| Mayflower Specialty / Embroker | Standalone AI liability | n/a | New entrants 2025–2026 |
Corgi COO Emily Yuan stated: "A lot of traditional carriers, they don't know how to underwrite AI and they're very spooked by AI." (Source: theinformation.com)
Insuring the data centers themselves
Distinct from covering losses caused by AI systems is covering the physical infrastructure that runs them. Swiss Re forecast, in a report published in September 2026 and reported on September 5, that global premiums for insuring data centers will likely reach $20 billion to $30 billion a year by 2030. The forecast was set against what the reporting described as a $1 trillion race to build data centers, with hyperscale facilities exposed to threats during both construction and operation (Source: wsj.com). Only the article's opening passage was retrievable; the Swiss Re report itself is not held, and no URL for it was given in the reporting. See AI Data Centers, Data Center Siting / AI Power Politics.
Risks and oversight
The AI LEAD Act (S. 2937) (proposed federal) and the Colorado AI Act (SB 24-205) and SB 25B-004 (Date Amendment) create liability exposure that insurers must price. Commentary has described an "insurer-as-regulator" dynamic seen historically in other risk areas: if AI-liability claims increase, the insurance market may play a structural role in AI governance through what it does and does not cover.
Federal proposals on AI claims denial
The Doctors Not AI Act of 2026, H.R. 10210, was introduced in the House on September 1, 2026 by Representative Greg Landsman (D-OH-1) with Representatives Buddy Carter (R-Ga.), Kim Schrier (D-Wash.) and Tom Barrett (R-Mich.). A bipartisan plan for such a bill had become public on September 2 through a paywalled trade account which described it as permitting insurers to use AI to process claims while requiring a human in the loop for denials based on medical necessity (Source: insideaipolicy.com).
The introduced text is narrower in one direction and stricter in another than that account. Its duty attaches only to adverse determinations, leaving approvals unregulated, and its trigger is wider than medical necessity alone — reaching determinations based on appropriateness, experimental or investigational exclusions, level of care, health care setting, effectiveness, clinical guidelines, utilization review criteria or generally accepted standards of care. What it permits is more constrained than a human in the loop: the licensed professional must conduct an independent evaluation of the individual's circumstances and "shall not treat any output of an artificial intelligence system as presumptively valid or defer to such output in lieu of independent clinical judgment" (H.R. 10210 — Doctors Not AI Act of 2026, as introduced). The oversight principle is the one at Human Oversight, applied at the point of the adverse determination rather than to the process as a whole.
The bill works through parallel amendments to the Public Health Service Act, ERISA and the Internal Revenue Code so that insured coverage, self-funded employer plans and the tax-code channel carry the same duty. Where an AI system was used, the denial notice must say so, describe the system's role and name the deciding professional with licence and credentials; the system, its role and its outputs enter the administrative record and are available to the enrollee on request. Its parity provisions make the use of an AI system in utilization review a treatment limitation for mental-health parity purposes and require plans to disclose AI use in the design or administration of any non-quantitative treatment limitation (H.R. 10210 — Doctors Not AI Act of 2026, as introduced). See Healthcare — AI Deployment.
Relationships
- deployed-by: Insurance carriers (claims, underwriting); Armilla (AI-risk coverage).
- regulated-by: State insurance regulators, NAIC.
- related: AI Liability, AI LEAD Act (S. 2937), Colorado AI Act (SB 24-205) and SB 25B-004 (Date Amendment), Enterprise AI Deployment Gap, AI and Cybersecurity (cyber-insurance overlap).
- related: Human Oversight, Healthcare — AI Deployment — the claims-denial oversight question.
- regulated-by: Doctors Not AI Act of 2026 (H.R. 10210) — proposed federal duty on AI in adverse benefit determinations.