The New AI Coverage Fight: Exclusions, Endorsements, and Denied Claims
Shumaker, Loop & Kendrick published a client alert on July 30, 2026, republished the same day by the National Law Review, under the heading “The New AI Coverage Fight: Exclusions, Endorsements, and Denied Claims.” Its argument is that insurers are no longer waiting for a claim to arrive before contesting coverage for AI losses. They are rewriting the forms now, through AI-specific exclusions, narrower endorsements, sublimits, and underwriting questions.
The exclusions you can see are the easy part. ISO form CG 40 47 01 26 strips bodily injury, property damage, and personal and advertising injury “arising out of, or attributable to,” generative AI from commercial general liability. Berkley has gone further, adding an “absolute” AI exclusion across several specialty liability lines. Broader wording is where the money is, because an exclusion reaching “the actual or alleged use, deployment, development, integration, or failure of AI” hands the carrier an argument whenever AI appears anywhere in the factual background. AI now sits inside office software, screening workflows, and call summaries.
“Silent AI” is going the way silent cyber went a decade ago. Until recently these losses were not treated as their own category: a cyber policy might respond to a breach involving an AI tool, a technology E&O policy to an AI-enabled service failure. The alert’s name for that unspoken coverage is “silent AI,” and its closing point is that insurers “are not waiting for AI coverage law to develop slowly through litigation.” The rules are being rewritten before the case law exists, which reverses the order in which the cyber market matured.
Negotiation, not litigation, is where this gets settled for now. The alert’s list of what to ask back is unusually specific: narrower causation language, carve-backs for ordinary business operations, an exception for AI used merely as a tool, and preserved coverage for cyber incidents, privacy claims, defense costs, and non-AI contributing causes. New wording is also the most negotiable wording, which is the argument for raising it this renewal rather than the next. How to read an individual form is worked through in our reading of ISO’s generative AI exclusion; the affirmative side of the same market is in our account of CFC’s AI coverage.
The cost of getting this wrong does not land on the risk manager who reads the endorsement. It lands on whoever assumed the CGL policy still worked the way it did last year: a mid-size firm whose chatbot said something actionable, an agency whose drafting assistant produced a summary a court later read as advice. The alert is candid that the insurer does not need AI to be the only cause, only that it was connected closely enough. Until a court puts a number on “closely enough,” every one of these exclusions is priced on a guess about how broadly it will eventually be read.