Insurance has become the most powerful de facto regulator of AI deployment in the United States, and it is currently regulating by withdrawal. On April 23, 2026, The Information reported that state insurance commissioners had quietly approved more than 80 percent of carrier requests to exclude AI-related damages from corporate insurance policies. This regulatory milestone attracted far less attention than major AI legislation, but experts argue it will likely matter more. For most American businesses, an activity that cannot be insured is effectively prohibited.

The insurance industry's retreat from AI coverage represents a structural problem, not a temporary market correction. According to a 2025 Geneva Association assessment, generative AI currently fails outright on three of the nine criteria that the insurance industry has used for over four decades to assess whether emerging risks are insurable. The most fundamental failure is information asymmetry: carriers cannot see which AI models their clients are running, how those systems are governed, or whether claimed risk controls actually exist. This makes it impossible for insurers to price two classical insurance market problems: moral hazard and adverse selection.

Why Does Insurance Matter More Than You Might Think?

The insurance industry has historically served as a powerful private-sector regulator for transformative technologies. Fire and building safety codes in the United States, for example, were originally drafted by fire insurance underwriters through the National Fire Protection Association, founded in 1896. These model codes carry no legal force until state and local governments adopt them, but governments almost always do because insurers reinforce them by grading and pricing how rigorously each community enforces them.

The Iranian oil sanctions of 2012 offer a vivid historical example of insurance's regulatory power. The United States and European allies prohibited most of the world's tanker ships from carrying Iranian crude oil not through military force, but by cutting off access to maritime protection and indemnity insurance. Without this insurance, oil tankers could not unload their cargo at major ports and were unwilling to accept new shipments. Iranian oil exports and government revenues plummeted as a result. More recently, when the Strait of Hormuz closed in 2026 as part of the U.S.-Iran conflict, insurance withdrawal was a major factor: P&I war-risk cover was removed on March 5, making the economic risk too high for shipowners to use the strait even if they could find willing sailors.

What Specific Insurability Problems Does AI Face?

The insurance industry uses what is known as the Berliner framework, a nine-criteria test for assessing whether emerging risks can be insured. Under this framework, generative AI currently fails on three criteria outright and is under strain on five more. Beyond information asymmetry, every other failure becomes easier to repair once that foundational problem is fixed. However, none of these problems will resolve quickly on their own. The cyber insurance market took roughly 20 years to mature, and it matured without ever developing the verification infrastructure that would have let underwriters discipline risk rather than merely price it. AI deployment is moving on a much faster clock, and the absence of insurance could pose a much greater adoption barrier.

• Information Asymmetry: Insurers cannot see which AI models clients are using, what applications they are deployed for, how governance structures work, or whether risk controls actually exist in practice.
• Loss-Frequency and Loss-Severity Data: The insurance industry lacks historical data on how often AI incidents occur and how severe their financial impacts are, making accurate pricing impossible.
• Verification Infrastructure: Unlike mature insurance markets, there is no established system for independent audits of AI risk controls, forcing insurers to guess rather than engineer their underwriting decisions.
• Correlated Risk: AI failures could affect multiple companies simultaneously in ways that exceed the private insurance market's capacity to absorb losses.

How Can Policymakers Fix the Insurance-AI Deadlock?

A comprehensive analysis from the Center for Strategic and International Studies (CSIS) proposes four specific recommendations, each tied to existing institutional templates and each aimed at a particular insurability failure. These steps are designed to convert insurance from an accidental brake on American AI adoption into an accelerator and one of the most effective safety regulators the United States has ever deployed.

• Build a Federal AI Incident Database: The federal government should establish a two-track AI incident database at the National Institute of Standards and Technology (NIST), modeled on aviation's Aviation Safety Reporting System. This is the cheapest recommendation and the one on which all others depend, as it repairs the missing loss-frequency and loss-severity data that all insurance pricing requires.
• Convene a Federal-State Working Group: The National Association of Insurance Commissioners should convene a federal-state working group before the current wave of AI exclusion templates hardens into a second generation of insurance forms that would be even more difficult to reverse.
• Prepare a Federal Catastrophic Loss Backstop: Congress should prepare, but stage, a layered, Price-Anderson-style federal backstop for catastrophic, correlated AI losses that the private market cannot absorb on its own.
• License Independent Verification Organizations: Governments should license independent verification organizations (IVOs) and accelerate the auditable risk-assessment standards they verify against. This pairing directly attacks information asymmetry and converts AI underwriting from guesswork into engineering.

None of these steps requires importing the European compliance model or the Chinese state-carrier model. Together, they represent a distinctly American approach to using market mechanisms and private-sector incentives to solve a public policy problem.

What Happens If Insurance Exclusions Become Permanent?

The stakes are extraordinarily high. While most enterprise liability policies currently are assumed to have blanket coverage that includes AI activities, insurers appear poised to exclude some or all AI uses from the next round of renewals, viewing AI risks as too broad, diverse, and obscure to price correctly. If this happens, AI adoption in the United States will undoubtedly slow as companies find AI adoption uninsurable and thus unattractive.

This dynamic makes AI insurance policy a form of AI industrial policy. The United States needs to take it seriously if it wants to maintain global leadership in the AI race. The insurance industry's current retreat from AI coverage is not a technical problem to be solved by better models or more data alone. It is a governance problem that requires coordinated action across federal and state regulators, Congress, and the insurance industry itself. Without that coordination, the invisible hand of the insurance market may prove far more effective at slowing AI adoption than any explicit regulatory prohibition.