If my business's AI chatbot or AI agent gives a customer wrong information, makes something up or takes a wrong action and someone sues, will my general liability, E&O, D&O or cyber insurance cover it, and what are the new 2026 generative AI exclusions?

What should I check, step by step?

1. List where your business uses AI in front of customers or in decisions: chatbots, AI agents that act or buy, AI-drafted advice, quotes, marketing copy, hiring screens. Note which vendor tools sit behind each.

2. At each renewal, search every policy (general liability, products, professional or tech E&O, D&O, EPLI, cyber, media) for "artificial intelligence", "generative", "algorithm" and "automated", including in definitions, application questions and endorsements, not only the exclusions page.
3. Look for the ISO form numbers CG 40 47, CG 40 48 and CG 35 08 on your general liability schedule of forms. Each one removes cover that wasn't specifically excluded before.
4. Ask your broker in writing how three claims would be handled and under which policy: a customer losing money after your chatbot gave a wrong answer, your AI publishing a false statement about someone, and an AI agent placing a wrong order or payment.
5. If you build or sell AI, check whether your tech E&O policy covers errors in AI outputs or carves them out, and whether it covers your contractual promises about accuracy.
6. Ask whether an affirmative AI endorsement or a standalone AI liability policy is available, and read what triggers it and what it excludes before comparing price.
7. Keep a record of each customer-facing AI answer or action (model version, the exact output or its hash, time) so you can show an insurer, a court or the customer what was actually said, rather than relying on a screenshot or a vendor log you can't reach later.
8. Tell your insurer about an AI incident on the policy's notice terms even if you think it's excluded; late notice can lose cover that would otherwise apply.

What do the ISO generative AI exclusions CG 40 47, CG 40 48 and CG 35 08 exclude?

Verisk's ISO, which writes standard policy forms used across the US market, made three generative AI exclusion endorsements available with an edition date of January 2026. All three define generative artificial intelligence 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.

• CG 40 47, for the commercial general liability coverage part, excludes bodily injury and property damage (Coverage A) and personal and advertising injury (Coverage B) arising out of generative AI. CG 40 48 excludes only Coverage B, the part that covers claims such as defamation and some advertising injury. CG 35 08 excludes bodily injury and property damage arising out of generative AI under the products and completed operations coverage part.
• The endorsements are optional underwriting tools, so whether one is on your policy depends on your insurer and state. A PropertyCasualty360 analysis of the forms notes that because this exposure wasn't specifically excluded before, attaching any of them will generally reduce coverage. A Gallagher Re paper (March 2026) adds that general liability doesn't cover first-party losses at all.

Which business insurance policy covers a claim caused by an AI chatbot or agent?

Insurance Journal (August 17, 2026) reported that carriers are increasingly filing with state insurance regulators to use the three ISO endorsements, with an insurance coverage lawyer expecting rapid adoption of some version of an AI exclusion in most lines; Verisk said it didn't know how many insurers had adopted them yet. The same article describes Berkley's absolute AI exclusion for directors and officers, errors and omissions and fiduciary liability products, which excludes claims based on use or development of AI, including content generated or disseminated using AI.

• A law firm review published June 15, 2026 describes the end of "silent AI", meaning AI risk that was covered because nothing excluded it. It reports some insurers declining to underwrite AI exposures, raising premiums, or carving out liability for AI outputs, decision-making and third-party AI tools, often through revised base forms, definitions and application questions rather than one obvious AI exclusion. It says cyber policies remain the most stable source of cover for AI-related cyber perils but weren't designed for liability from AI outputs.
• The same review explains that coverage is usually decided by how the claim is framed, not by the technology: a software failure may fall under tech E&O, discrimination by an AI hiring tool under EPLI, confidential data put into an AI tool under cyber or privacy cover, and AI-generated content infringing IP under media, tech E&O or general liability. That applies to businesses that only use vendor AI tools, not just AI companies.

Is there insurance that covers AI hallucinations and AI agent mistakes?

Some insurers now sell affirmative AI cover, which names AI failures as covered instead of relying on silence. One example is a Lloyd's coverholder whose AI liability policy, launched in 2025 with Lloyd's underwriters, is triggered by AI underperformance, including errors, hallucinations or inaccuracies leading to damages, and whose published coverage areas include third-party financial loss from inaccurate outputs, claims from AI agent mistakes and defamation from outputs. Exclusions and limits apply, and the products are new.

• Affirmative endorsements on existing policies can also narrow cover by protecting only the scenarios they list, the June 2026 review warns. Compare the trigger, the definitions of AI and of a claim, retentions and sublimits against the claims you listed with your broker.
• Every one of these policies still needs proof. Whether the claim is an AI error, an advertising injury or a professional mistake, the insurer and the other side will ask what the system actually output, which version produced it, and whether a human approved it before it went out.