Does my business insurance cover an AI mistake? CG 40 47 and the generative-AI exclusion, for a 3-person business

A small business that lets AI write to customers may no longer have general liability cover for what that AI says. Since 1 January 2026, ISO has offered three optional endorsements, CG 40 47, CG 40 48 and CG 35 08, that can remove claims "arising out of" generative AI from a commercial general liability policy, according to MoneyGeek.¹ PYMNTS reported in May 2026 that state regulators had approved more than 80% of AI-exclusion filings.²

The question for a three-person business is narrow. Does your policy carry one of those form numbers, and what would an AI misquote or refund promise cost you if it did? This piece separates what is settled from what is reported, and ends with three questions for your broker. It is not legal or insurance advice; check with your broker and counsel.

What changed on 1 January 2026?#

ISO, the insurance industry's standard-forms body, introduced three optional generative-AI exclusion endorsements effective 1 January 2026, as MoneyGeek reports them.¹ An endorsement is a page added to a policy that changes its coverage. These three are optional: an insurer must file them, and then choose to attach them to a given policy.

Source: MoneyGeek, 2026.¹ Valet has not reviewed the ISO form text itself; read the endorsement attached to your own policy.

The definition carries most of the weight. MoneyGeek reports that the forms define generative AI 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."¹ A chatbot fits that definition. So does an assistant that drafts or sends customer emails.

Is the exclusion in force, and on whose policy?#

The exclusion is available, not universal. Its legal status differs by policy, and three facts determine whether it applies to you.

1. The form exists and is effective. MoneyGeek gives 1 January 2026 as the effective date.¹
2. Your insurer has filed it in your state. PYMNTS reported in May 2026 that regulators had approved more than 80% of AI-exclusion filings, with Florida, Connecticut and Maryland seeing the highest volume.² That is a reported industry figure, not a list of which insurers attached which form.
3. Your policy carries it. This is the only step that settles the question for you. It appears, if at all, on the declarations page or the forms and endorsements schedule.

PYMNTS names Berkshire Hathaway, Chubb, Travelers and Berkley in its report on large insurers pulling back from AI risk.² Do not read that as proof that any particular policy from those carriers excludes AI. Read your own schedule.

Does it apply to a small business like mine?#

The generative-AI exclusion has no size threshold in the reporting reviewed here. Applicability runs through three gates, and a business is in scope only if it passes all three.

Gate 1: an ISO-based general liability policy#

Start here if you carry general liability, on its own or inside a business owner's policy. MoneyGeek reports that ISO forms underpin about 82% of US property and casualty policies.¹ If your policy uses ISO wording, the exclusion is at least available to your insurer.

Gate 2: the endorsement is on your schedule#

Search the forms schedule for "CG 40 47," "CG 40 48" or "CG 35 08." An insurer may also use its own wording rather than ISO's, so also look for "generative," "artificial intelligence" or "AI" in endorsement titles. If you renewed after 1 January 2026, redline the new schedule against last year's.

Gate 3: your AI talks to customers#

The exposure exists where AI output reaches someone outside the business. A website chatbot, an assistant that answers inquiry emails, an AI that writes quotes or confirmations: each produces "content or responses" a customer acts on. AI used only for internal drafts that a person rewrites before sending creates far less of that exposure.

PYMNTS put the size question directly: "Small to mid-sized firms, often without specialized coverage, face the greatest exposure."²

What AI mistakes could a small business face?#

The AI mistakes a small business is most likely to face are small and contractual, judging by the public cases: a wrong price, an invented policy, an unauthorised discount. The table below lists them with the closest public example. Whether any of these losses was ever covered by general liability, even before the exclusion, is a question for your broker; do not assume it was.

Liability for these sits with the business. In the Air Canada ruling, the tribunal rejected the argument that "the chatbot is a separate legal entity that is responsible for its own actions."⁴ PYMNTS summarised the trend in July 2026: companies remain liable for what their AI systems say, including when the system is "filling a knowledge gap."⁵ No disclaimer on a chat widget is a safe harbor. No court case involving a business of three people was found, but nothing in these rulings depends on size.

What must, should and may a small business do?#

The obligations around AI and insurance divide into three tiers. "Must" marks what follows from your policy terms or settled case outcomes. "Should" marks prudent practice. "May" marks options.

What does dedicated AI insurance cost, and is it sized for you?#

Dedicated AI insurance exists, but its 2026 pricing is built for larger firms. PYMNTS reported in May 2026 that specialist carriers and startups, including Corgi, Armilla, Mayflower Specialty, Embroker and Munich Re, write AI cover with limits of $2 million to $50 million and premiums from "hundreds to several hundred thousand dollars annually."² In September 2026, AIUC raised a $40 million Series A to certify and insure AI agents.³

PYMNTS also quoted the industry's own framing: "Insurers are not saying AI is uninsurable. They are saying they do not yet know what it costs."² For a three-person business, the materiality threshold matters more than the limit. A misquoted package or an honoured discount is likely to fall in the hundreds or low thousands of dollars. Compare that against your deductible before you shop for a $2 million policy. That comparison is our reading of the public cases, not an actuarial figure.

Three questions to ask your broker#

The three broker questions below turn the exclusion from a headline into a fact about your policy. Send them by email so the answer is written down.

Subject: AI and our general liability policy

Hi [broker name],

We use an AI assistant to [answer customer emails / run a website chat / draft quotes]. Before our renewal on [date], could you answer three questions in writing?

1. Does our current policy include CG 40 47, CG 40 48, CG 35 08, or any other endorsement that excludes or limits claims involving generative AI or artificial intelligence? If yes, which form, and since which date?

2. If our AI assistant told a customer a wrong price, promised a refund we do not offer, or double-booked a date, would any part of the resulting claim be covered, under this policy or any other we hold? Was that true before the exclusion as well?

3. Will our renewal add any AI exclusion, and if so, what would it cost to remove it or to add cover for this risk?

Thanks,
[name]

What is still unclear#

Three points about the generative-AI exclusion remain unsettled as of October 2026. First, the scope of "arising out of" in an AI context has no court interpretation in the sources reviewed. Second, the 80% approval figure is a reported industry total, not a count of policies that carry the form.² Third, this piece relies on MoneyGeek's description of the forms rather than the ISO text.¹ Treat each as a question for your broker, not a conclusion.

For the controls that reduce what an AI can get wrong in the first place, see the six controls for AI customer replies.

Where Valet fits#

Valet is not an insurer, and Valet makes no claim that it is insured or that it changes your coverage. What it does is narrow what an AI can say on your behalf. Valet answers a small business's written inquiries only from the owner's own facts and never invents prices, discounts or policies. Money questions, complaints, contracts and deposits go to the owner, and the customer is told when the owner will reply. Every reply carries a Valet receipt: what was asked, the rule used, the sources and where it was sent from. The free lead check reviews 10 to 20 recent inquiries and shows what went unanswered.