TL;DR: ISO/Verisk made three optional generative AI exclusion endorsements, CG 40 47, CG 40 48, and CG 35 08, available in January 2026 for commercial general liability and products/completed operations policies. The language is broad enough to bar coverage for AI-related claims even when AI is only a contributing factor. W.R. Berkley introduced an absolute AI exclusion for its D&O, E&O, and fiduciary lines. Small teams should check every renewal for this language rather than assume it is not there. New affirmative coverage exists too, including HSB AI Liability Insurance for small and mid-size businesses, announced March 2026.
For most of 2025, small teams treated "does our insurance cover AI risk" as a hypothetical. It stopped being one in January 2026, when Verisk's ISO division made three optional endorsement forms available: CG 40 47 01 26 and CG 40 48 01 26 for commercial general liability (CGL) policies, and CG 35 08 01 26 for products/completed operations policies. Each one excludes coverage for losses arising out of generative AI. ISO forms are the standard wording many US insurers build their commercial policies on, and carriers have been filing with state regulators to use these three, so when ISO ships a new exclusion, it does not stay theoretical for long, it shows up at renewal.
If your team runs an AI product, uses AI tools in client-facing work, or has ever had a chatbot, AI-generated report, or AI-assisted decision touch a customer, this is not a compliance footnote. It is a coverage gap that can show up the moment you actually need to file a claim.
What the ISO endorsements actually exclude
The three forms attach to different coverage parts, so the practical effect depends on which one your carrier uses.
| Form | Applies to | What it removes |
|---|---|---|
| CG 40 47 01 26 | Coverage A and Coverage B | Bodily injury, property damage, and personal/advertising injury "arising out of" generative AI |
| CG 40 48 01 26 | Coverage B only | Personal and advertising injury arising out of generative AI (narrower than CG 40 47) |
| CG 35 08 01 26 | Products/Completed Operations Liability Coverage Part (Section I) | Bodily injury and property damage arising out of generative AI |
All three define "generative artificial intelligence" the same way: 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. That definition is wide enough to cover a customer-facing chatbot, an AI writing tool used in marketing copy, or an AI coding assistant that shipped a bug into a product your business sold.
The phrase doing the most damage to policyholders is "arising out of." In exclusions it is usually read broadly, as some causal connection rather than a direct cause, though the exact test varies by state. A claim does not need to be primarily about AI to trigger the exclusion, AI just needs to be somewhere in the chain of events. That means a defamation claim over AI-drafted marketing content, an IP infringement claim over AI-generated images, or a product liability claim where an AI recommendation played any role, can all potentially fall outside coverage.
Because the ISO forms are optional, individual carriers decide whether and how to attach them. A carrier can attach the broad CG 40 47, use the narrower CG 40 48 (or CG 35 08 on a products/completed operations policy), or write proprietary exclusion language instead of adopting the ISO form outright. There is no single public registry of which carrier is using which version, which is exactly why asking directly at renewal matters more than assuming based on what you read about the industry in general.
It is not just general liability
The CGL exclusion got the endorsement numbers, but it is not the only place this is happening. Tech E&O, professional liability, and cyber carriers have been following the same pattern, sometimes with even broader language.
W.R. Berkley introduced an "absolute" AI exclusion in 2025 for its D&O, E&O, and fiduciary liability products, Insurance Journal reported in August 2026. It excludes claims tied to the use or development of AI, including content generated or disseminated with AI, the insured's own AI policies and procedures, and breaches of duty in using AI, not just generative AI specifically. That is a meaningfully wider exclusion than the CGL forms, and it is worth checking for by name if your team carries D&O or E&O coverage.
Coverage-by-coverage checklist
Use this table at your next renewal. For each policy your team holds, confirm the status directly with your broker rather than assuming based on last year's terms.
| Policy type | Common exposure if AI-related | What to check for | What to ask for |
|---|---|---|---|
| General liability (CGL) | Bodily injury, property damage, or advertising injury tied to an AI-generated output | CG 40 47, CG 40 48, or CG 35 08 attached at renewal | A copy of the exact endorsement, and whether a carve-back or buy-back option exists |
| Tech E&O / professional liability | Errors from an AI-assisted deliverable, AI coding tool, or AI-generated advice | Any exclusion referencing "artificial intelligence," "machine learning," or "algorithmic" outputs | Whether AI-related errors are still covered under the core professional services definition |
| D&O | Claims that board or executive decisions relied on flawed AI output or governance failures | Absolute AI exclusion language, similar to what W.R. Berkley introduced | Confirmation of whether AI governance failures are treated differently from AI performance failures |
| Cyber | A breach or incident where an AI tool was the entry point or contributing cause | Whether the cyber policy's AI carve-outs overlap with your GL exclusion, creating a gap between both | An affirmative statement that AI-involved incidents remain covered under the cyber policy |
| Dedicated AI liability | Model failure, hallucination-driven claims, algorithmic bias | Whether this coverage exists at all, most small teams do not have it yet | A quote for a standalone AI liability rider or policy, even if you do not buy it this cycle |
Coverage that did not exist a year ago
The exclusion wave has a mirror image: insurers are also building affirmative AI coverage, because the exclusions created demand for something to fill the gap. Three options are relevant depending on your size.
HSB AI Liability Insurance, from Munich Re's HSB division, was announced on March 18, 2026, specifically for small and mid-size businesses. It is built to pay for AI-related losses that some GL policies exclude, including bodily injury, property damage, and advertising injury claims tied to AI-generated advertising, marketing copy, blog content, or social media. Standard limits are $25,000 or $50,000 with a $500 deductible, and higher limits are available. HSB does not sell it directly: pending approval by insurance regulators, it is added to the business policies of partner carriers, so ask your broker whether your carrier offers it.
Munich Re's aiSure program, offered through a February 2026 partnership with Mosaic Insurance, provides up to 15 million (EUR, USD or CAD) in initial capacity aimed at AI developers and vendors rather than general AI users. It is a performance model: coverage responds when an AI model fails to meet clearly defined performance thresholds. This is the right tier if your business builds or resells an AI product, not if you are simply a user of AI tools.
Vanguard AI, a February 2026 structure from Armilla and Chaucer at Lloyd's, bundles $25 million or more in dedicated AI liability limits with $10 million in cyber limits. This is enterprise-scale coverage, most small teams will not need it, but it is useful context if you are evaluating an AI vendor's own insurance posture during due diligence.
For most small teams, the realistic near-term move is asking your existing carrier whether HSB's small-business product, or an equivalent from a competing insurer, is available as an add-on. It is a meaningfully cheaper starting point than a dedicated $15 million policy built for AI vendors.
The renewal checklist, copy this into your next broker email
Before you sign any renewal for GL, Tech E&O, D&O, or cyber coverage in 2026, send your broker these questions in writing so you have a documented answer:
- Has an AI exclusion endorsement, ISO form or proprietary, been added to this policy at this renewal?
- If yes, can you send the exact endorsement language, not just a summary?
- Does the exclusion use "arising out of" or a narrower "solely caused by" standard? The difference matters enormously for how easily a claim gets denied.
- Does our current policy cover claims where an AI tool was used by an employee, even if AI was not the primary cause?
- Is an affirmative AI liability rider, buy-back endorsement, or standalone policy available through this carrier?
- If we use AI tools from a third-party vendor, does a claim against us get treated differently than a claim against the vendor directly?
- Has this carrier filed or adopted CG 40 47, CG 40 48, or CG 35 08 in our state, and was that filing approved by the state regulator?
Get the answers in writing. Broker conversations are useful, but only the policy language and the written confirmation from the carrier hold up if a claim is ever denied.
Self-audit before you even call your broker
- List every AI touchpoint in your business, not just your product. Marketing copy, coding tools, customer support chatbots, and internal AI-assisted decisions all count as potential exposure, even if AI is not your core business.
- Pull your last two renewal declarations pages and compare the endorsement list. A new exclusion form appearing between renewal cycles is the clearest signal something changed.
- Check whether your vendor contracts shift liability back to you. If an AI vendor's terms disclaim their own liability for output errors, and your insurer excludes AI claims, you can end up holding both ends of the risk with no coverage at all.
- Separate "AI as your product" from "AI as a tool you use." The coverage gap and the available fixes differ significantly between the two, and most small teams are the latter, which is exactly the segment HSB's small-business product targets.
- Put a date on this. AI exclusion adoption is moving fast enough in 2026 that a policy reviewed six months ago is not a reliable guide to what is in force today.
What we could not confirm: Verisk does not publish the endorsement text openly, so the form descriptions above come from Verisk's own descriptions as reported by the Big "I" and Insurance Journal. We did not review state case law on "arising out of"; ask coverage counsel how your state reads it.
Correction, October 22, 2026: an earlier version said the Mosaic and aiSure partnership dates from March 2026 (Mosaic announced it on February 26, 2026), said aiSure pays without a negligence finding (Mosaic describes cover for failing defined performance thresholds), said W.R. Berkley "confirmed" its absolute exclusion (Insurance Journal reports it introduced the exclusion and could not reach Berkley for comment), and said the ISO forms took effect on January 1, 2026 for all CGL policies (they are optional forms with a January 2026 edition date, and CG 35 08 is for the products/completed operations coverage part) (Mosaic, Insurance Journal).
Sources: Big "I", Verisk to roll out new general liability exclusions for generative AI exposures, Insurance Journal, August 17, 2026, HSB press release, March 18, 2026, HSB AI Liability product sheet, Mosaic press release, February 26, 2026, Chaucer press release on Vanguard AI.
Related Reading
- AI liability insurance in 2026, what coverage actually exists
- AI vendor due diligence checklist, 30 questions before you sign
- AI vendor contract red flags, 12 clauses that create liability
- GenAI vendor risk assessment framework for 2026
- Third-party AI tool risk assessment template
- VC AI governance due diligence, the 18-item checklist
- AI spend governance, token budget controls that stop runaway bills
