Guide
AI exclusions in liability insurance: what changed
Liability insurance was written before AI agents did real work. Insurers are now adding exclusions that remove AI-related losses from standard policies, so a business can no longer assume its existing cover will respond when AI gets something wrong.
Two terms first
An exclusion is the part of a policy that lists causes or conditions it does not cover. An endorsement is a form attached to a policy that adds, removes or changes coverage. Most AI exclusions arrive as endorsements added to an existing policy at renewal.
What ISO filed for general liability
ISO, the Verisk unit that writes standard policy forms many US insurers use, made a multistate general liability filing in July 2025, proposed to take effect January 1, 2026. Verisk described it as adding “new, optional endorsements designed to address GenAI liability exposures.”
Trade press reporting identifies three endorsements:
| Form | What it excludes |
|---|---|
| CG 40 47 | Bodily injury, property damage, and personal and advertising injury arising out of generative AI, in commercial general liability |
| CG 40 48 | Personal and advertising injury arising out of generative AI, in commercial general liability |
| CG 35 08 | Bodily injury and property damage arising out of generative AI, in products and completed operations coverage |
These are optional. Each insurer decides whether to attach them, so two businesses with the same kind of policy can end up with different AI coverage.
What carriers have done
Some carriers have gone further than the ISO forms. Insurance Journal reported in July 2026 that W. R. Berkley introduced an “absolute” AI exclusion for directors and officers, errors and omissions, and fiduciary liability products. Filings with state regulators show what an insurer may use, not necessarily what it attaches to every policy.
What “silent AI” means
“Silent AI” describes policies that neither clearly cover nor clearly exclude AI-related losses. The term borrows from “silent cyber,” the same question the market worked through for cyber risk. Silent exposure is uncomfortable for insurers because one AI failure can cause many similar losses at once, across many policyholders. Explicit exclusions are one way insurers are resolving it.
What it does not mean
The NAIC’s 2023 model bulletin on artificial intelligence is about how insurers themselves use AI in underwriting, pricing and claims. It is not about excluding AI risk from the policies they sell. The two are often confused.
What this means for businesses using AI vendors
- Read the renewal. Look for new endorsements that mention artificial intelligence, generative AI or automated systems.
- Check the vendor contract’s limitation of liability: what it caps, and what kinds of loss it excludes.
- Ask both sides directly: if the AI gets this wrong and it costs us money, who pays?
For the terms used here, see the glossary. For how a warranty differs from a guarantee and from insurance, see warranty vs guarantee vs insurance.
Sources
- Verisk: Emerging risks in ISO General Liability multistate filing (July 2025)
- Claims Journal: ISO generative AI exclusion endorsements (July 20, 2026)
- Insurance Journal: W. R. Berkley AI exclusion (July 22, 2026)
- NAIC: Model Bulletin on the Use of Artificial Intelligence Systems by Insurers
- NAIC: what is an insurance endorsement or rider?
- California Department of Insurance glossary