The AI Exclusion Clause Now Has a Form Number. Does Your Policy Have One?
For most of this year, "AI exclusions are coming" has been a conversation about direction, not specifics. That changed in the last two weeks of July. According to reporting from Carrier Management and Claims Journal, insurers are actively evaluating and filing new ISO generative-AI exclusion endorsements, with named form numbers now attached to the trend: CG 40 47, CG 40 48, and CG 35 08, all commercial general-liability and products-completed-operations endorsements (Carrier Management; Claims Journal).
Lathrop GPM partner Alana McMullin told both outlets she expects "rapid adoption of some version of an AI exclusion" across "most lines of coverage." Verisk vice president Joe Lam framed the underwriting logic behind it plainly: "everyone acknowledges this is new technology that is going to create new exposures." Neither outlet's reporting establishes how many carriers have actually filed or adopted the forms yet. That's the detail worth sitting with. The exclusion language has gone from theoretical to specific enough to name, without yet being universal enough to assume.
That's exactly the gap that turns a compliance footnote into a live coverage question. If your renewal conversation last spring covered AI exclusions as a "watch and wait" item, this is the point where watching stops being sufficient, because the industry is no longer debating whether the language exists. It's debating how far it spreads and how fast. A board or risk committee that keeps asking "will AI exclusions happen" is asking last quarter's question. The current one is narrower and answerable today: what does our policy, our specific endorsement schedule, actually say right now?
The number worth remembering is 978%: a Gallagher study using Testudo Global tracking data found AI-related lawsuits increased that much from 2021 to 2025. Carriers aren't filing exclusion language in a vacuum. They're responding to a litigation trend that's already showing up in the loss data, which is exactly why this is arriving through underwriting and actuarial judgment rather than one dramatic headline event. There's no single incident to react to here. There's policy language quietly narrowing.
This month's action: pull your commercial general liability and products-completed-operations policies and check the endorsement schedule specifically for CG 40 47, CG 40 48, and CG 35 08, along with any carrier-specific manuscript language that isn't an ISO form at all. Then do the same for E&O, cyber, D&O, and any other management-liability policy separately. Those CGL form numbers don't govern those lines, and each one needs its own read for AI-related definitions, exclusions, conditions, and limitations. Bring your broker three questions in writing, not a general "are we covered?" that produces a general "probably": has any carrier we use filed or proposed AI exclusion language in our states, and what exact language applies to our policy; how would each of our actual policies respond to a claim arising from an AI failure; and have we reviewed our AI-vendor contracts for indemnification language that actually protects us if the vendor's tool fails. Get the answers in writing. A verbal "you should be fine" isn't a coverage position, and it won't control when a claim arrives.