Insurance Companies Are Excluding AI Risk. That's Not an Opportunity.

The companies deploying AI want insurance. The companies that sell insurance don't want the risk.

That's the actual state of AI liability coverage — and it runs in the opposite direction from the headlines about surging demand making insurance stocks attractive.

Over the past year, some of the largest U.S. property and casualty insurers have done something you can usually trust them to do: protect their own book. Berkshire Hathaway, Chubb, Travelers, AIG, and W.R. Berkley filed requests to exclude AI-related damages from standard commercial liability policies. More than 80% of those filings were approved by state regulators. The exclusions began taking effect in early 2026.

This isn't a story of insurers expanding their books to capture a growing market. It's a story of carriers that looked at AI risk and decided it was a problem they couldn't price.

Why insurers can't price it

Insurance works on a simple principle that goes back decades: the risks in your book should be independent of each other. One homeowner's fire shouldn't cause another's to burn. That independence lets the law of large numbers smooth out the unpredictable, and the insurer keeps the difference between expected losses and collected premiums.

AI violates that principle. A single defect in a foundation model — say, a widely used AI agent begins giving harmful advice — can cascade across thousands of insured companies simultaneously. The industry calls this "accumulation risk." McKinsey put it bluntly: AI liability and systemic cyber risk may be "correlated in ways that make them genuinely difficult to diversify and price responsibly".

A 2025 assessment by the Geneva Association ran generative AI through the industry's standard nine-criteria test for insurability and found three outright failures — maximum possible loss, loss frequency, and information asymmetry — and only one pass. Insurers can't verify what models their policyholders are using, how those models are governed, or whether risk controls exist. As one analysis put it, insurers are treating AI like a used car lot where they can't tell the good ones from the bad ones.

The litigation environment confirms their anxiety. U.S. generative AI lawsuits grew 978% between 2021 and 2025, according to a reinsurance broker's count. Another 137% year-over-year jump followed in 2024-25.

Faced with that, carriers did what carriers do when they can't measure the downside: they stopped writing the risk.

What the major carriers actually did

The Insurance Services Office — which underpins roughly 82% of U.S. property and casualty policies — introduced optional AI exclusion endorsements effective January 2026, covering bodily injury, property damage, and advertising injury tied to generative AI. A census of nearly 10,000 filings in July 2026 found more than 60 property and casualty carrier groups had filed on AI exclusions: 41 groups filed to adopt exclusions and 20 groups filed to delay adoption.

Some exclusions are narrow. Others aren't. W.R. Berkley introduced an "absolute" AI exclusion that can bar claims involving the use, deployment, development, or failure of AI — not just AI output. It names tools like ChatGPT explicitly.

The effect is to strip AI coverage from the default policies that most commercial customers buy. If you're a company using AI for customer service, hiring, marketing, or any other function, your standard general liability, errors and omissions, or directors and officers policy may now explicitly exclude the risks that AI creates. The coverage gap sits with the business, not the insurer.

The "opportunity" that isn't a stock thesis

The market response to the exclusions has been the creation of standalone AI liability products — mirroring what happened with cyber insurance roughly two decades ago. Munich Re launched aiSure. Lloyd's syndicates and startups like Corgi, Armilla, and Testudo offer policies with limits typically between $2 million and $50 million.

Projections say the standalone AI insurance market could grow from about $40 million in 2024 to roughly $4.8 billion by 2032, at a compound annual growth rate near 80%. Those are the numbers that fuel the "AI insurance opportunity" narrative.

But here's the constraint that matters for a stock thesis: even at $4.8 billion, AI insurance is expected to account for only about 0.34% of commercial property and casualty premiums by 2032. The growth rate sounds explosive only because the base is tiny.

More importantly, the companies most likely to benefit from this niche are not the publicly traded insurers that retail investors watch. Munich Re — which has the earliest and most developed AI-specific offering — is traded in Frankfurt, and the aiSure revenue stream is a rounding error in a multi-hundred-billion-euro consolidated book. The specialty providers like Corgi and Armilla are private startups. Hiscox, which acquired Corix (Vouch Insurance's underwriting division), is listed in London. None of the major U.S. insurers that retail investors buy — Chubb, Travelers, Berkshire Hathaway — are positioning as AI liability specialists. They're positioning as carriers that have avoided the problem.

What this means for the stocks you can actually buy

Chubb (NYSE: CB) trades at a market cap of $130 billion with a trailing P/E of about 11.7 and a 15.5% return on equity. It generates roughly $60 billion in annual revenue, with nearly half from international markets. Travelers (NYSE: TRV) has a $76 billion market cap, a trailing P/E near 9.2, and 93% U.S. revenue concentration. Berkshire Hathaway (NYSE: BRK.B) sits at $1.1 trillion, its insurance subsidiaries embedded in a conglomerate where no single line can be isolated.

None of these companies will move their earnings meaningfully from AI liability premiums. If anything, the exclusions are defensive moves that protect their existing books from an unmeasured tail risk. The economic consequence for these stocks isn't a growth story — it's a risk management one. These carriers are choosing to exclude what they can't price, rather than risk a correlated loss event that could devastate a book they've built over decades.

That's the kind of decision that makes a property and casualty insurer worth owning, but it doesn't create a rerating catalyst.

The real investment question

The AI insurance narrative asks the wrong question. It treats AI as a new premium stream that flows into insurers' pockets. The evidence suggests the flow goes the other way: from the public's expectation of coverage, into the gap between what businesses need and what carriers will write.

The bottleneck behind this story isn't demand. Businesses deploying AI want coverage and would pay for it — over 90% of corporate insurance decision-makers across six major markets say they need AI-tailored coverage, and two-thirds would pay at least 10% more in premiums for it. The bottleneck is supply, and specifically the structural problem that makes AI hard to insure. Until someone solves the information asymmetry — how do you verify an insured's AI governance? — or the correlation problem — how do you cap a loss that can hit every policyholder at once? — the market stays in exclusion mode.

A CSIS report from September 2026 went so far as to suggest a Price-Anderson-style federal backstop for catastrophic AI losses, the same framework that enabled the nuclear industry to obtain insurance. That's a signal of how severe the private market views the problem.

For the investor, the takeaway is this: the major U.S. insurance stocks aren't playing offense on AI. They're building walls. The companies that might eventually profit from AI-specific coverage are private, tiny, or too diversified for it to matter. The gap between the headline and the structure is wide enough that it's worth closing before you buy the narrative.