What your policy quietly stopped covering
For most companies the assumption was simple: whatever the business does, the general liability, directors-and-officers, and professional indemnity policies stand behind it. That assumption is being rewritten in the fine print. The Insurance Services Office, whose standardized wording underpins much of the market, has published two optional endorsements, CG 40 47 and CG 40 48, that allow insurers to carve generative AI out of standard commercial general liability cover. Once an insurer attaches that language, a claim arising from an AI output simply falls outside the policy.
This is not theoretical. Carriers including Chubb, Travelers, and Berkshire Hathaway have secured regulatory approval to add explicit AI exclusions across general liability, D&O, and errors-and-omissions lines, with some taking effect from January 2026. The exposure does not announce itself. It arrives as a clause at renewal, and for most boards the first sign that their AI risk is uninsured will be a claim that gets declined.
Why the gap lands hardest on the company using AI
The instinct is to assume the AI vendor carries the risk. The structure of the market says otherwise. Technology errors-and-omissions cover is designed for the supplier of a technology, not for the enterprise deploying it, so a company running a third-party model holds the legal responsibility for its outputs while holding none of the cover built for that risk. Cyber policies, meanwhile, were never written for hallucinations that cause financial loss or disclose data, and product-liability cover routinely excludes purely algorithmic failure.
The pressure behind the exclusions is real. Generative-AI lawsuits grew roughly 978 percent between 2021 and 2025, with more than 700 cumulative cases and a 137 percent jump in the most recent year alone. Insurers are repricing a risk they cannot yet model, and the cleanest way to do that is to exclude it. Specialist cover does exist, from Munich Re's aiSure product to newer carriers such as Armilla and Testudo, but it is capped, conditional, and tied to ongoing model-quality assessments most companies have never run.
What a serious company should check now
The first task is not to buy more insurance. It is to read the policy you already hold. Establish whether your current general liability, D&O, and professional indemnity wordings contain an AI exclusion, and whether one is being introduced at the next renewal. That single question, put to your broker before renewal rather than after a claim, is what separates a managed exposure from a discovered one.
From there the work is to map where AI actually creates liability in the business, which outputs carry real financial or reputational cost, and what evidence of control a specialist insurer would demand before writing cover at all. The companies that treat insurability as a board-level question now will keep using AI with a known risk position. The ones that wait will learn the hard way that adopting the technology and insuring it were never the same decision.
Read next: Your AI Agent Can Now Spend Your Money. The Hard Part Is Everything That Happens After. · Europe Is Building Its Own Frontier AI Model. That Changes the Choice You Thought You Had.