The Insurance Industry Doesn't Get to Opt Out of AI
Three major carriers filed in November 2025 to add absolute AI exclusions to their commercial policies, and your clients' AI exposure keeps growing anyway. This is a broker problem. Carriers walking away are forcing you to sell a policy with a hole in it. Carriers that AREN'T walking away already have affirmative AI products on the market. The broker's job is to know the difference and steer the client toward the second group.
The Retreat
AIG and Great American filed generative AI exclusions with US regulators in November 2025, with W.R. Berkley taking the most aggressive line. Berkley's "absolute AI exclusion" bars claims tied to any actual or alleged use of AI. It reaches models the insured doesn't own, third-party tools, licensed software, and AI features embedded in vendor products. That's nearly every modern business application. AIG told the Financial Times it has no plans to implement the exclusions yet, but the form is filed and ready to deploy. Great American took a narrower path, updating its Washington commercial umbrella and excess forms to exclude generative AI liability. The direction is the same. When losses come in, the carriers want the exit ramp already paved.
The Problem Is Your Client Is Still Liable
The Moffatt v. Air Canada decision in February 2024 settled the question brokers should be bringing to every renewal. Air Canada's chatbot told a customer about a bereavement fare policy that didn't exist. The customer relied on it and sued. She won. The tribunal rejected outright Air Canada's claim that the chatbot was a separate entity. The company that deployed the AI was liable for what the AI said.
That ruling is the template for AI liability now. It doesn't matter whether your client built the model or licensed it from a vendor who promised it was safe. When the AI hallucinates something a customer relies on, the enterprise using it pays. A carrier with an absolute AI exclusion writes a policy that covers none of that. Your client will discover the gap the first time it matters.
The Market Is Already Correcting
The carriers filing absolute exclusions aren't the whole market. Munich Re's aiSure product covers hallucinations and performance failures across generative and traditional ML models, and is model-agnostic by design. Earlier in 2025, Google announced a partnership with Beazley and Munich Re to build AI-specific cyber coverage for enterprise deployments, with Chubb joining the same framework. On April 30, 2025, Armilla launched an AI liability policy underwritten at Lloyd's by Chaucer, with limits up to $25 million per insured. That policy explicitly covers hallucinations, model drift, inaccurate outputs, data leakage, and AI regulatory violations.
So the industry hasn't decided AI is uninsurable. Part of it has decided it doesn't want to do the work. The rest of it is quietly building the book the first group is walking away from.
The Counter-Argument
The carriers adding exclusions aren't irrational. Model outputs are opaque and loss history is thin. The plaintiffs' bar is early on AI liability theory. When Chaucer's Tom Graham says "we will be selective, like any other insurance company", he's conceding that affirmative AI coverage still requires underwriting discipline. A carrier declining AI exposure may be pricing the uncertainty honestly.
What that argument doesn't defend is absolute exclusion language. An absolute exclusion isn't underwriting discipline. It's an abdication of the business. Insurance prices uncertainty. That is the product. A carrier that refuses to price a widespread commercial exposure because the math is hard is telling its brokers to find another market.
What to Do at the Next Renewal
Read every commercial liability policy for AI exclusion language. Look specifically for "absolute" and for exclusions that reach third-party or embedded AI. Flag them before renewal. Then price the alternative: Armilla's affirmative product at Lloyd's is accessible through specialty brokers and works as primary or excess. Munich Re's aiSure runs through its industrial client channel. Chubb and Beazley are live in the Google partnership model.
The conversation with a skeptical client is simpler than it looks. Ask how many AI tools the team already uses and whether there is any written AI governance around them. Then ask what happens if their chatbot does what Air Canada's did. If the answer is "we assume our E&O covers it" and the policy has Berkley's absolute exclusion, the real answer is "we pay out of pocket." That's not a risk most clients want on their balance sheet.
Carriers don't get to watch the AI era arrive and then refuse to price the consequences. Brokers who understand which markets are still doing the work will win the accounts of every client that figured out AI isn't going away.