ISO's Generative AI Exclusion Is Now Attaching to General Liability Renewals
On July 17, 2025, Verisk’s ISO General Liability team filed a multistate update giving carriers a way to take generative AI losses out of commercial general liability. The endorsements are optional, and the filing proposed a January 1, 2026 effective date. The forms carry a 01 26 edition date, and a Big “I” Virtual University summary reports them effective in January 2026, according to Verisk. CG 40 47 removes bodily injury, property damage, and personal and advertising injury arising out of generative AI from both Coverage A and Coverage B. CG 40 48 does the Coverage B half alone. CG 35 08 does the same work inside products and completed operations.
What the filing was is the tell. Generative AI appeared in it as one of four emerging risks, sitting alongside human trafficking, social inflation, and cyber war, under filing designations GL-2025-OFR25 for forms and GL-2025-ORU25 for rules. AI did not get its own process. It was productized through the same emerging-risk machinery that handles everything else, which is how silent cover actually gets switched off: not by announcement, but by a form number showing up in a renewal package.
The seam to watch is the definition. CG 40 48 defines generative artificial intelligence as “a machine-based learning system or model that is trained on data with the ability to create content or responses, including but not limited to text, images, audio, video or code.” That describes producing content. It does not obviously describe taking actions, and taking actions is what the systems insureds are standing up this year do. W. R. Berkley drew its perimeter differently and earlier. Its absolute AI exclusion for D&O, E&O, and fiduciary lines, circulating by May 2025, defines AI by inference rather than by generation, and reaches any use, deployment, or development of it.
Two exclusions, two perimeters, and a category of exposure sitting in between. Verisk appears to be looking at that gap already: The Insurer reported on July 10, 2026 that ISO is weighing exclusions aimed specifically at agentic AI.
The renewal question is narrow and answerable. Pull the endorsement schedule, find whether an AI exclusion is attached, and read its definition rather than its title. A form drafted around generation may leave an agentic deployment covered; a form drafted around inference probably does not. Several broker summaries now circulating on these forms describe them as excluding AI generally, which is broader than the ISO wording reads. What a company falls back on once cover narrows is its own documentation, and the program that produces it is laid out in our guide to AI governance in insurance.