CG 35 08: The ISO Exclusion – Generative Artificial Intelligence (Products and Completed Operations)

By Joel R. Singh · Last verified: 2026-07-29 · How we verify

What CG 35 08 is and what it does to your policy

A standard commercial general liability policy actually contains more than one coverage part. The part most people know covers premises and ongoing operations: slip-and-fall, a contractor who accidentally breaks a water pipe while on the job, a vendor whose employee causes damage at a client site. A separate coverage part, Products and Completed Operations, covers harm that arises after the work is done or after the product leaves your hands. A customer is injured by equipment you sold them six months ago. A software application you delivered causes data loss at a client three weeks after go-live. That post-delivery exposure is what Products and Completed Operations covers, and CG 35 08 carves generative AI out of it entirely.

ISO (now operating under Verisk) filed three generative AI exclusion endorsements simultaneously for a January 1, 2026 effective date. CG 40 47 attaches to the standard CGL Coverage Part and excludes both bodily injury/property damage and personal and advertising injury arising from generative AI. CG 40 48 is a narrower companion, covering only Coverage B (personal and advertising injury) on the standard CGL form. CG 35 08 sits apart: it attaches specifically to the standalone Products and Completed Operations Liability Coverage Part, not to the standard CGL form, and it excludes Coverage A (bodily injury and property damage) in that context. A business that carries the Products and Completed Operations coverage as a bundled section of a full CGL policy needs to confirm whether that section can also receive CG 35 08, and a business that carries the coverage as a standalone part should treat CG 35 08 as its primary concern.

The practical effect is that if your carrier has adopted CG 35 08 and you ship products or deliver finished work that incorporates generative AI, the post-delivery liability coverage on your CGL program no longer responds to bodily injury or property damage claims with a generative AI nexus. The gap exists from the moment the policy renews with the endorsement attached.

The operative exclusion language

The ISO form text for CG 35 08 01 26 is a subscriber-only ISO Commercial Lines Manual document; verbatim quotation requires direct access to that filing. Based on consistent reporting from the IIABA (Big I) and confirmed by multiple trade sources including Insurance Journal and the Olshan Frome Wolosky analysis published in The Legal Intelligencer (May 2026), the endorsement excludes coverage for:

The definition of "generative artificial intelligence" in the endorsement is reported 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." Verisk is reported to have drawn this definition from existing government and NAIC frameworks, making it unlikely to face state filing objections on novelty grounds. The "including but not limited to" language keeps the definition open-ended beyond the enumerated output types.

Two drafting choices deserve attention. First, "arising out of" is one of the broadest causal phrases in insurance contract law. Courts have consistently interpreted it to require only a minimal nexus between a loss and the triggering condition, not direct or proximate cause. A claim where generative AI played a background role, not the central role, can still trigger the exclusion under this standard. Second, the definition covers any system trained on data that can generate output: not only large language models, but image generators, code synthesis tools, AI video systems, and audio generation models. A business that uses generative AI as one component of a larger product, rather than as the core product, should not assume the exclusion is inapplicable.

Source for exclusion language and definition: IIABA / Big I, "Verisk to Roll Out New GL Exclusions for Generative AI Exposures"; cross-checked against Crawford, The Legal Intelligencer / Olshan Law, May 2026 and Testudo glossary, "Generative AI Exclusion".

When the exclusion actually bites: two scenarios

Scenario 1: AI customer service chatbot leaks PII in a delivered SaaS product. A software company ships a SaaS platform to a mid-size healthcare client. The platform includes a generative AI chatbot for patient-facing scheduling. Three months after go-live, the chatbot surfaces protected health information from one patient's record in a response to a different patient. The harmed patient brings a claim for personal injury (privacy invasion) and associated financial harm. The SaaS company's CGL products and completed operations section would ordinarily be a candidate for response, since the harm arose from a completed, delivered product. With CG 35 08 attached, the carrier denies the bodily injury and property damage component of the claim because it arises out of the generative AI component of the delivered software. Coverage B personal and advertising injury (privacy-invasion claims) is a separate question governed by the standard CGL form and CG 40 47, not CG 35 08, but CG 35 08 has already removed the Coverage A path. The company is left to look at tech E&O, cyber, or a standalone AI liability policy for any response.

Scenario 2: AI-generated engineering specification causes property damage after project completion. An engineering consultancy delivers a project specification that includes AI-generated material calculations. The client begins construction; a structural component fails because the AI-generated load estimate contained an error not caught in review. The resulting property damage claim is brought months after the consultancy's engagement ended, squarely in completed-operations territory. With CG 35 08 on the policy, the CGL products and completed operations coverage cannot respond to property damage arising out of generative AI outputs embedded in the delivered work product. The consultancy's professional liability (E&O) policy becomes the primary potential coverage path, but only if that policy contains affirmative AI language or does not itself exclude AI-generated output. See Does E&O Cover AI? for the gap analysis.

How CG 35 08 interacts with other coverage and what fills the gap

CG 35 08 operates only within the Products and Completed Operations coverage context. It does not affect Coverage B personal and advertising injury on the standard CGL form (that is CG 40 47's domain), and it does not affect tech E&O, cyber, D&O, or professional liability policies, which follow their own forms and underwriting. The coverage gap created by CG 35 08 is therefore bounded but specific: post-delivery bodily injury and property damage with a generative AI nexus. Several coverage paths exist to address it.

Tech E&O with affirmative AI language. A professional liability policy covering technology services can respond to completed-work claims, including AI-output errors, if the policy contains affirmative AI coverage. Many standard tech E&O forms are silent on AI or contain their own exclusions; review the policy form, not just the marketing description. The E&O gap page tracks which carriers have issued affirmative AI language in their tech E&O forms.

Standalone AI liability products. Several specialty carriers have designed products explicitly to fill the gap left by the January 2026 CGL exclusions. Relm Insurance's PONTAAI is structured as an excess difference-in-conditions (DIC) wrap policy, designed to respond where existing liability programs have excluded AI-related losses. It covers bodily injury and property damage from AI system failures, IP infringement, discrimination, privacy breaches, professional negligence, and AI regulatory violations. Testudo, backed by Lloyd's syndicates including Apollo Underwriting and QBE, launched a generative AI liability policy in January 2026 explicitly targeting the gap created by the ISO exclusion forms. Armilla AI offers standalone AI liability coverage underwritten through Lloyd's syndicates including Chaucer, covering hallucinations, model drift, inaccurate outputs, data leakage, and regulatory violations, with limits reported up to $25 million per organization. Vouch provides AI-related errors, discrimination, IP, and regulatory defense as part of its tech-startup coverage package. The full verified carrier list is on the Carrier Comparison Table.

Carrier-specific affirmative AI endorsements. Some carriers that have adopted CG 35 08 also offer a buy-back endorsement to restore coverage for qualified insureds, typically conditioned on AI governance documentation: a model inventory, usage policy, and evidence of human review. Ask your broker specifically whether a buy-back is available from your current carrier before shopping replacement coverage.

For the full matrix of which forms exclude what, and which carriers have adopted each ISO form versus a proprietary variant, see the Exclusion Tracker. For the E&O interaction specifically, see Does E&O Cover AI?.

Buyer and broker checklist

Adoption timeline. The three ISO AI exclusion forms (CG 35 08, CG 40 47, CG 40 48) carry a January 1, 2026 effective date. Adoption varies by carrier and by state, since state insurance departments must approve each filing. Carriers including W.R. Berkley, Chubb, Travelers, Berkshire Hathaway, and Cincinnati Financial had filed to adopt the Verisk forms or proprietary equivalents by approximately April 2026, per trade press coverage. Not every policy renewing after January 1, 2026 automatically carries these forms; adoption depends on whether your specific carrier filed and received approval in your state, and whether the renewal documents reflect the filing. The form number in your endorsement schedule is the only reliable check, not the renewal date alone.

Last verified: 2026-07-29. Source links checked on that date. Report an error.