The Wheel of New Coverage

How Insurance Turns Ambiguity Into Exclusions, Sublimits, and Eventually a Product

By James W. Moore

Key Takeaways

• New exposures often arrive under old policies before insurers have agreed what those policies were supposed to cover. The resulting uncertainty can produce exclusions, limits and new products.
• A narrower coverage grant does not make an unfamiliar risk predictable. It can make the insurer’s promise manageable enough to underwrite deliberately, collect experience and reconsider its terms.
• Mold, pollution and cyber illustrate different outcomes: bounded buybacks, specialized protection, and a large market of standalone and packaged products. There is no guarantee that every exposure follows the same path.
• AI is beginning to show both sides of the process: exclusions and uncertainty in established lines alongside affirmative specialty coverage. The real question is which exposure insurers will define and assume next.

The Industry Discovers the Future. Again.

I have watched the insurance industry discover the future several times. It usually begins by denying that the old policy was ever intended to cover it.

That sounds a little cynical, but think about how often it has happened. Water damage turned into a mold controversy. Pollution claims turned familiar liability language into years of litigation. Computer failures and data breaches tested contracts written before cyber liability had a name.

The underlying hazard was not always new. What changed was the amount of money involved, the number of people affected, or the possibility that an existing insurance policy might respond. Once that happens, insurers face an awkward question: What did we actually agree to cover?

The answer tends to set off a familiar sequence. The industry restricts an exposure, describes it more precisely, and eventually offers some portion of the risk back to buyers willing to pay for it. I think of it as the wheel of new coverage. Not every risk makes a complete turn, and some take a different exit along the way. But the mechanism keeps reappearing.

Drawing a Smaller Box

The first move is often to remove an uncertain exposure from a broad policy form. That understandably frustrates policyholders. It can also look like the insurer has decided the risk is uninsurable, although the decision may be considerably narrower.

Consider an insurer that has no reliable way to estimate a potentially open-ended loss under its existing contract. It might still consider an endorsement covering a clearly defined event, subject to a separate premium, a sublimit, and underwriting conditions. The insurer has not suddenly learned the true frequency or severity of the risk. It has limited what it promises to pay under that particular coverage, then decided whether the premium, available capital, and expected accumulation justify taking the remaining uncertainty.

A $100,000 limit caps one contract’s response. It doesn’t mean the insurer can safely issue a million such contracts without considering whether the same event could trigger thousands of them. And a known maximum claim does not tell us how often claims will happen.

What more specific wording does provide is a way to learn. When coverage is clearly defined, insurers can begin comparing claims arising under similar terms. That doesn’t make the risk predictable overnight, but it gives underwriters something more useful than arguing over what an old policy was supposed to cover. Limits, exclusions, deductibles, risk controls and reinsurance make it possible to evaluate narrower commitments while that experience develops.

Sometimes experience supports wider coverage. Sometimes the insurer tightens the terms or exits the risk. The important development isn’t simply the exclusion. It’s what the market is prepared to offer afterward.

This is why I have always been skeptical of the word uninsurable when it is used without qualification. “Uninsurable” often means not insurable on the old form, at the old limit, for the old price.

Mold: The Buyback With Boundaries

Texas gave us a particularly clear example in the early 2000s. Melinda Ballard’s water-damage claim became a major mold dispute with Fire Insurance Exchange. A Texas jury returned an award exceeding $32 million in 2001. On appeal, the court reversed significant portions of the award while upholding approximately $4 million in actual damages. It was the kind of case that made everyone pay attention, but the numbers behind the broader market problem were already alarming.

In its 2001 mold coverage order, the Texas Department of Insurance reported that mold claim frequency in its study had increased more than sixfold in 18 months, while the average mold claim cost approximately $18,000. Existing homeowners rates had not been set with that recent loss experience in mind. Insurers were pulling back from particular homeowners forms and asking to restrict coverage. The department had to balance the cost of claims against continued availability of homeowners insurance.

The initial proposal included a flat $5,000 basic mold limit, plus an opportunity to buy more. Both consumer and insurance-industry commenters objected to the flat limit, for different reasons. The commissioner rejected it. Instead, the adopted forms preserved defined coverage for ensuing mold from otherwise covered water damage, while excluding additional remediation and testing costs beyond specified covered repairs.

Insurers were also required to offer additional mold coverage for a separate premium. Buyers could choose coverage tied to 25, 50 or 100 percent of specified underlying policy limits, subject to a maximum for the policy period. Insurers could decline a particular buyback on sound underwriting grounds related to the exposure. The revised endorsements became available on January 1, 2002, with mandatory implementation required by January 1, 2003.

There’s the wheel in miniature, with an important twist. The regulator didn’t simply accept the smallest proposed box. The final boundaries came out of a dispute about what protection homeowners actually needed and what the market could afford to provide.

It was not a return to unlimited mold coverage. Nor was it a declaration that no mold loss could ever be insured. A contentious exposure became a more explicit insurance transaction.

Pollution: Sometimes the Coverage Moves

Pollution demonstrates a less comforting version of the same process.

For years, insurers and policyholders fought over the meaning of the pollution exclusion’s “sudden and accidental” language. Did sudden mean abrupt? Could gradual contamination qualify as accidental? The answers had enormous financial consequences. In Morton International v. General Accident (1993), the New Jersey Supreme Court found that the exclusion had been presented to regulators as a clarification even though it sharply reduced coverage, and declined to enforce it as written in that dispute.

The industry’s response included changing the contract language. In 1985, ISO explained a broader pollution exclusion and, crucially, described ways to purchase certain excluded protection separately. The later decision in Kimber Petroleum v. Travelers records those alternatives. One endorsement deleted the pollution exclusion except for government-ordered cleanup costs. Two separately structured pollution-liability coverage parts offered differing treatment of cleanup expenses.

Pause over that: the broader exclusion and the routes to alternative coverage were described as part of the same redesign. The market wasn’t necessarily waiting years to discover a way of insuring pollution. It was deciding that ordinary general liability coverage was the wrong place to leave certain losses, while creating other ways to underwrite them.

Some exposures don’t return to the form where the dispute began. They move to a specialty product, with a different underwriting process, price and set of terms. That’s another possible turn of the wheel.

Cyber: A Market Can Grow Before the Language Catches Up

Cyber took another path. By 2020, the U.S. cyber insurance market generated roughly $4.1 billion in premium, according to NAIC reporting. About $2.58 billion was standalone cyber coverage, while approximately $1.49 billion was written as part of package policies. That is a real market, although the loss ratios reported for major writers varied considerably. Developing a product and fully understanding its losses are different achievements.

The revealing evidence comes from Lloyd’s. In July 2019, it issued Market Bulletin Y5258, requiring managing agents to clarify whether policies affirmed or excluded cyber-related losses. The first phase covered specified first-party property policies incepting on or after January 1, 2020.

Consider the timing. Cyber insurers were already selling substantial affirmative coverage, but other policies could still contain cyber exposure nobody had clearly intended to insure or exclude. The product existed before the silence was resolved.

Lloyd’s wasn’t inventing cyber insurance with that bulletin. It was requiring its market to become more explicit about promises embedded, or potentially embedded, in traditional contracts.

AI: Where Does the Buyback Go?

We are beginning to see the same argument unfold around artificial intelligence.

RAND’s September 2026 report, The Insurability of Artificial Intelligence, describes a mixed response: some affirmative coverage, some new exclusions, and substantial uncertainty where existing policies do not explicitly address AI-related losses. RAND proposes an AI Coverage Notice to identify what is affirmed, excluded or left silent. The resemblance to Lloyd’s silent-cyber initiative is hard to miss.

AI also presents a more diffuse problem. Cyber became a recognizable insurance category, even as it continued to overlap other lines. AI is embedded in activities insured under commercial general liability, technology errors and omissions, professional liability, cyber, directors and officers, and elsewhere. A claim arising from an AI-generated recommendation may look quite different from one involving an autonomous physical system. There isn’t one neat “AI policy” into which every exposure fits.

Verisk’s account of its ISO general liability multistate filing identifies generative AI as an emerging exposure addressed in the filing; a Gallagher analysis discusses the optional exclusion endorsements addressing different parts of commercial general liability. Their development is evidence of an effort to clarify contractual boundaries. How often carriers actually attach them is a separate question, and the forms alone don’t answer it.

The other side of the wheel is already visible. In April 2025, Chaucer and Armilla AI announced standalone third-party liability protection for specified AI-model failures. Their February 2026 Vanguard AI announcement describes cyber and technology E&O coverages operating alongside a separate AI-liability policy, with rules for allocating mixed claims and dedicated AI aggregate limits of $25 million or more. That tells us how the insurers propose to divide the risk. It does not tell us whether buyers will adopt the structure widely or whether the pricing will prove adequate.

The resemblance to pollution in the mid-1980s is intriguing. There, broader exclusions and separate options for coverage were developing together. Today’s AI market likewise shows both efforts to constrain broad forms and efforts to insure more specific losses elsewhere. That’s a comparison of how the market is organizing coverage, not a prediction that AI will remain a specialty risk for decades.

One obstacle will be accumulation. If a thousand policyholders rely on the same AI model, cloud provider or upstream system, losses may arrive together. A $100,000 sublimit at each insured doesn’t make a thousand simultaneous claims trivial. Insurers need to recognize common dependencies and decide how much aggregate exposure they can support.

Not every AI risk shares the same problem. A closely monitored model-performance guarantee may behave quite differently from a widely deployed autonomous platform. Sorting out those differences is exactly what underwriting and product development are for.

The Next Turn

Mold, pollution and cyber arrived at different coverage structures. Nothing says AI must follow any one of them.

What does look familiar is the industry’s effort to move from ambiguous promises toward defined ones. Some exposures will be excluded, some insured on limited terms, and others handled through specialist coverage. Eventually, somebody will announce the latest turn as a breakthrough product.

Meanwhile, instead of repeating the question of whether AI is insurable in the abstract, we can ask something more useful.

Which piece of the risk will the market be willing to sell back next?

Sources

• Texas Department of Insurance, Mold and Other Fungi Mandatory Amendments Order CO-01-1105
• Texas Department of Insurance, Mold, Fungi or Other Microbes Endorsements
• Texas Court of Appeals, Allison / Fire Insurance Exchange v. Ballard
• Kimber Petroleum Corp. v. Travelers Indemnity Co. (1997)
• NAIC, 2020 cyber insurance premium report
• Lloyd’s, Market Bulletin Y5258 (July 2019)
• RAND, The Insurability of Artificial Intelligence (September 2026)
• Verisk, Emerging Risks in ISO General Liability Multistate Filing (2025)
• Gallagher, ISO Introduces Generative AI Exclusion in Commercial General Liability Policies
• Chaucer, Armilla Standalone AI Liability Product (April 2025)
• Chaucer, Vanguard AI Coordinated Coverage (February 2026)

AI Disclaimer: This content was created with assistance from artificial intelligence technology. While content is based on factual information from the source material, readers should verify all details directly with the respective sources before making business decisions.