Your next commercial lines renewal may show something you haven't seen before: endorsements that strip out coverage for losses "arising out of" generative artificial intelligence. Verisk's ISO unit rolled out standardized AI exclusion forms effective January 1, 2026. Carriers that subscribe to ISO forms can now attach them to CGL and products/completed operations policies with a single filing. If you use generative AI anywhere in your operations—and at this point most companies do—you need to know what those forms do, where else AI exclusions are landing, and how to negotiate before the policy is bound.
What actually went live in January 2026
Verisk released three optional endorsements with a January 2026 edition date. They're optional in the sense that each carrier chooses whether to use them; once a carrier files them, they become part of the standard offering unless you negotiate otherwise.
CG 40 47 applies to the ISO Commercial General Liability Coverage Part (occurrence or claims-made). It excludes coverage under Coverage A and Coverage B for bodily injury, property damage, or personal and advertising injury arising out of generative artificial intelligence. CG 40 48 narrows the focus to Coverage B only—personal and advertising injury arising out of generative AI. CG 35 08 applies to the ISO Products/Completed Operations Liability Coverage Part and excludes Section I coverage for bodily injury or property damage arising out of generative AI.
The definition of "generative artificial intelligence" in these forms is worth reading verbatim. Verisk defines it 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 language is drawn from government and NAIC-style definitions. It's broad enough to capture ChatGPT, Midjourney, DALL·E, code-generation tools, and most other systems that generate net-new output from training data. It does not limit itself to a specific vendor or use case. If your marketing team uses an AI copy tool, your product embeds a generative feature, or your customer support uses a chatbot, a claim that "arises out of" that use could fall within the exclusion—depending on how your carrier applies the form and how a court would read it.
Why Carriers Wanted These Forms
Insurers have been asking for underwriting tools to address generative AI for a simple reason: the loss environment is uncertain and the litigation is already real. As of late 2025 there were at least 11 major U.S. lawsuits involving generative AI, from copyright and IP claims to harmful chatbot interactions. In one tragic case, the parents of a child who died by suicide sued, alleging that information furnished by a generative AI chatbot contributed to the death. In professional services, lawyers have been sanctioned for submitting briefs that cited fake cases hallucinated by ChatGPT. Bain reported that 95% of U.S. companies use generative AI; insurers are not wrong to treat it as a pervasive exposure. Some carriers and commentators cite studies suggesting that general-purpose generative AI tools produce inaccurate outputs at rates in the 58–88% range—a statistic that underwriters use to justify exclusions or sublimits rather than leaving the exposure fully in.
The ISO forms give carriers a standard, filed way to exclude that exposure instead of drafting one-off manuscript exclusions. That doesn't mean every carrier will attach them on every policy. It does mean that the option is now in the toolkit, and early signals suggest strong interest. Verisk has said it expects insurers to roll these out quickly. The question is whether they'll appear on yours, and in what form.
The "Absolute" AI Exclusion and Where Else Exclusions Are Showing Up
The ISO forms are modular: a carrier can use CG 40 47 (full CGL AI exclusion), CG 40 48 (Coverage B only), or neither. Some carriers have gone further. Berkley, for example, has deployed what commentators call an "absolute" AI exclusion—language that excludes coverage for any actual or alleged use, deployment, or development of artificial intelligence, including the generation, creation, or dissemination of content or communications using AI. That's broader than the ISO definition and broader than "generative" AI; it can sweep in uses that don't involve generative systems at all. Surplus lines carriers have been using AI exclusions for a while; the January 2026 ISO rollout brings the same type of risk transfer into the admitted market in a standardized way.
Don't assume AI exclusions will stay in CGL or products. Carriers are filing and using AI exclusions in professional liability (E&O) and D&O as well. Design professionals, tech companies, and others have seen AI exclusions added to E&O forms; D&O policies are also a target. When you get your renewal, check the full package—CGL, products, E&O, D&O, cyber, media—for new AI-related endorsements. The hollowing out of coverage can happen across the program, not just on one line.
Negotiate at Renewal, Not After a Loss
The single most important move is to negotiate the scope of any AI exclusion at placement or renewal, before a claim occurs. Once a loss happens, the policy language is fixed. Before binding, you have leverage: the carrier wants the premium and the relationship. After a claim, you're arguing about interpretation.
Where the carrier insists on an AI exclusion, three levers matter.
Definitions. If the form uses "artificial intelligence" or "generative artificial intelligence," push for a clear, bounded definition. The ISO definition is at least written down; manuscript forms sometimes use "AI" or "machine learning" without definition, which invites disputes over whether a given system is in or out. Narrow the term if you can—e.g., to systems that generate content or make autonomous decisions in the relevant context—so that routine or tangential use doesn't trigger the exclusion.
Lead-in language. Many exclusions bar coverage for claims "arising out of," "based upon," or "in any way connected to" AI. The last of those is the most dangerous: "in any way connected to" can support a denial where the only link to AI is that someone used an AI tool somewhere in the chain (e.g., to draft marketing copy for a product that later caused injury). Push for causation language that requires the claim to arise from the AI output or the AI's use in a meaningful way, not just a remote connection. Replacing "in any way connected to" with "arising out of" is an improvement; "caused by" or "directly and proximately caused by" is better still if the carrier will agree.
Carve-backs and endorsements. If you can't remove the exclusion, negotiate carve-backs for defined, lower-risk uses. For example: AI used only for internal drafting or research, not customer-facing output; or AI tools that are vetted and used in accordance with a documented policy. Some carriers will add a limited carve-back for specified operations or a sublimit instead of a full exclusion. Your ability to get that depends on your ability to describe your AI footprint clearly—which is why the next step matters.
Map Your AI Footprint Before the Underwriter Asks
Underwriters are asking more pointed questions about where and how insureds use AI. If you can't answer, you risk a broad exclusion by default, or worse: a rescission or denial later on the basis that you didn't disclose a material exposure. So before renewal, map where AI is used across the organization—product, marketing, customer support, HR, legal, operations. Document who uses it, for what, and whether the output is customer-facing or internal. That map does two things: it supports a more precise carve-back (e.g., "we use AI only for internal drafting, not for client deliverables"), and it gives you accurate answers for the application. Inaccuracies or vague answers can be used against you if a claim is denied.
Engaging coverage counsel at placement is increasingly common for companies with meaningful AI use. Counsel can help draft carve-back language, compare forms across carriers, and advise on how courts are likely to interpret broad lead-in language if a dispute arises.
If You Get a Denial Anyway
Exclusionary language is construed against the drafter in most jurisdictions. Courts often require that an exclusion be clear and that it not swallow the grant of coverage. An "absolute" or very broad AI exclusion may be fertile ground for dispute where the connection between the claim and AI is tenuous—for example, a professional malpractice claim where the only AI touchpoint was the use of an AI tool to create marketing materials for the same firm. Some policyholders have successfully argued that enforcing a broad exclusion would defeat the reasonable expectations of the insured. So if you're faced with a denial based on an AI exclusion, don't assume the carrier's reading is final. Have coverage counsel analyze the wording, the facts of the claim, and the applicable law. Insurers can overreach, especially with new, untested exclusions.
Filling the Gap: Standalone and Specialist Products
As AI exclusions spread, specialty products are emerging. Lloyd's coverholders and other carriers are offering standalone generative AI liability policies that cover errors, IP infringement, defamation, bodily injury, property damage, and data disclosure arising from AI outputs. Those products don't replace a thoughtful negotiation of your primary and excess program, but they can provide a dedicated limit for AI-related claims where your standard CGL, E&O, or D&O has been cut back. If you've negotiated carve-backs and still have meaningful residual exposure, a specialist policy may be worth pricing.
Verisk's January 2026 forms are a turning point: they normalize AI exclusions in the standard market. The carriers that adopt them are not doing so in isolation; they're responding to real litigation and real uncertainty. Your job at renewal is to make sure the exclusion you get is as narrow as your risk profile and your footprint justify—and to know exactly where you stand before the next claim lands.