General liability is the policy that answers when a third party says your product or your words hurt them. That is where a chatbot harm claim lands, and CG 40 47 removes bodily injury, property damage and personal and advertising injury from Coverages A and B when the claim arises out of generative artificial intelligence [3]. The definition in the forms is written around what the system does, not where it sits: a machine-based learning system or model trained on data with the ability to create content or responses, including text, images, audio, video or code [6]. Verisk says it built that wording from existing government and NAIC definitions [7].
The drafting itself, not anyone's forecast, sets the consequences here. The three forms together close off the general liability and products/completed operations routes for the same class of loss: Coverages A and B under CG 40 47, Coverage B alone under CG 40 48, and Section I products liability under CG 35 08 [3][4][5][15]. And because the definition names code, the excluded class reaches coding assistants inside the software pipeline, not only customer-facing text and images [16].
This matters at scale because of how many companies are exposed. Bain reports 95% of U.S. companies currently use generative AI [10]. That is 19 in 20 [14], against a form that a carrier can attach at renewal to any commercial general liability policy on the ISO base [3].
The filing and the form language are public record [1][2]. How many carriers actually adopt the exclusions is not: Verisk says it has already received considerable carrier interest and expects insurers to roll the exclusions out quickly, and that expectation is Verisk's own, reported by IndependentAgent.com, with no count of carriers attached [8][18]. The endorsements are optional, so nothing forces a carrier to use them [3].
The ceiling is higher than the ISO forms. Berkley has already deployed an absolute AI exclusion covering any actual or alleged use, deployment or development of artificial intelligence by any person or entity, including the generation, creation or dissemination of any content using AI [13]. Surplus lines carriers moved on AI exclusions before the ISO filing [12]. Meanwhile the claims that prompted the underwriting response are live: at least 11 major U.S. suits spanning copyright infringement to harmful chatbot interactions, including a suit filed last year by the parents of a child who died by suicide, alleging information from a generative AI chatbot contributed to the death [9][11].
The endorsement numbers show up on the schedule of forms on a renewal quote, which is the cheapest place to learn whether a carrier attached one and whether any affirmative product replaces what it removes [19].
