Buying AI Liability Insurance: Questions to Ask Your Broker and Insurer

The questions to ask before buying AI liability insurance fall into four groups: what the policy treats as an AI claim; how its limits, defense costs and reporting rules work; who stands behind it; and how it fits with the D&O, EPL and E&O policies the company already holds. A broker should be able to answer each one from the policy wording rather than the marketing, because two AI policies with the same limit can respond to very different claims.

What Should You Prepare Before Talking to a Broker?

The most useful preparation is a short account of where the company uses AI and who could bring a claim over it, because a broker can only test a quote against exposures it knows about. There are four groups of claimants to consider, and each tends to point to a different policy.

Each of these groups has already brought claims over AI. Customers and clients sue over wrong output; in Moffatt v. Air Canada, decided on February 14th 2024, a British Columbia tribunal held the airline liable for wrong bereavement-fare advice from its website chatbot and rejected the argument that the chatbot was responsible for its own actions ^[1].

Job applicants and employees bring discrimination claims; in Mobley v. Workday, filed on February 21st 2023, a federal court in California let disparate-impact claims (claims that an apparently neutral practice disproportionately harmed a protected group) proceed against the vendor of an AI screening tool, and preliminarily certified a nationwide collective of applicants aged 40 and over ^[2][3]. In September 2026, more than three years after it was filed, the case was at the class certification stage ^[4].

Shareholders bring securities claims and derivative claims (suits brought on the company's behalf against its directors and officers), and investors filed 15 AI-related securities class actions in the first half of 2026, a pace that Cornerstone Research said would nearly double the 2025 total ^[5]. Regulators bring enforcement actions, such as the SEC's settled charges of March 18th 2024 against two investment advisers, Delphia and Global Predictions, over misleading statements about their use of AI, with penalties of $225,000 and $175,000 respectively ^[6].

It is best to write this map down, with the AI systems behind each exposure, before the first conversation with a broker.

Which AI Claims Will the Policy Cover?

An AI policy's definitions and exclusions decide which claims it covers more than its title does, so five questions about scope should be put to the broker for every quote.

1. The definition of AI: A definition tied to generative AI can leave out the scoring, screening and pricing models behind many decisions about people. The generative AI exclusions for general liability that ISO (the Verisk unit that publishes standard US policy forms) announced in October 2025 define generative AI as a machine-based learning system or model trained on data that can create content or responses such as text, images, audio, video or code ^[7]. A policy built on a similar definition could leave a hiring algorithm outside it, so the definition should match the company's AI inventory.
2. Users as well as developers: A company can be held responsible for what its AI tells customers, as the Air Canada decision shows, so a policy that excludes AI supplied by third parties would miss much of the exposure of a company that uses vendors' models (a deployer, in the EU AI Act's terms).
3. Employment, discrimination and securities claims: A policy written for claims over AI output may exclude these, yet they are the claims D&O and EPL policies exist for, and the cases above show they are already being brought over AI. The broker should show where the wording brings them into cover.
4. Regulatory proceedings: Several new AI laws are enforced by regulators rather than private plaintiffs; Texas's Responsible AI Governance Act, in effect since January 1st 2026, is enforced by the state attorney general, with civil penalties of up to $200,000 for each violation that cannot be cured ^[8]. The broker should confirm whether investigation costs are covered and whether fines are covered where the law allows them to be insured.
5. Territory: The EU AI Act reaches providers and deployers outside the EU when their AI system's output is used in the EU, and its highest fines reach €35 million or 7% of worldwide annual turnover ^[9]. A company with customers or staff abroad should therefore ask whether the policy responds to claims and proceedings outside the United States.

These answers should come from the specimen wording and endorsements sent with each quote (the sample policy form and the amendments to it), because a quote summary rarely shows the definitions and exclusions that decide whether an AI claim is covered.

Should You Buy a Standalone AI Policy, AI Modules or a DIC Layer?

The right form of AI cover depends on who could sue and how sound the existing program is; for a company exposed to applicants, shareholders or regulators as well as customers, that points to affirmative AI modules or a difference-in-conditions (DIC) layer rather than a standalone policy. Affirmative cover means the AI risk is granted expressly in the wording rather than left to argument.

Mayflower Specialty writes the second and third forms: AI Directors and Officers Liability (AI-D&O), AI Employment Practices Liability (AI-EPL) and AI Professional Liability (AI-E&O) modules, plus AI DIC Excess over an existing tower, meaning the stack of primary and excess policies a company already holds. The coverage overview explains how they fit together, and the choice of form should follow from the claimant map.

How Do Limits, Defense Costs and Reporting Rules Work?

Limits, defense costs and reporting rules decide whether an AI claim reaches the policy at all and how much of the limit is left when it does: a sublimit or defense costs inside the limit reduce what is available to pay, and a late report can put a claim outside cover. The glossary defines these and other policy terms.

Sublimits and Defense Costs

Mayflower's application states that defense costs reduce and may exhaust the limit of liability and are applied against the retention, which is the amount the company pays toward a claim before the policy responds, much like a deductible. AI disputes can run for years, as the Workday litigation shows, so the limit should be sized for a long defense as well as a settlement. The broker should also confirm the size of the retention and whether one AI failure affecting many people counts as a single claim, since that can decide whether one retention applies or many.

Claims Made and Reported, Retroactive Dates and Extended Reporting

A claims made policy needs only the claim to be made during the policy period, while a claims made and reported policy also requires it to be reported in time, although more generous versions allow a window of 30 to 60 days after expiration ^[12]. Mayflower writes its coverage on a claims made and reported form, so every demand letter, regulatory inquiry or AI incident that could become a claim should be reported promptly.

Because many companies have used AI for years, the retroactive date should match the expiring D&O, EPL and E&O policies where possible, and Mayflower's application asks for the current program and a proposed retroactive date for this purpose.

The broker should confirm whether an ERP is available, for how long and at what cost before the policy is bound (put into force).

Who Stands Behind an AI Liability Policy?

The insurer that issues an AI liability policy stands behind it, and its financial strength matters because liability claims can be paid years after the premium. One widely used measure of that strength is AM Best's Financial Strength Rating, an independent opinion of an insurer's ability to meet its ongoing insurance policy and contract obligations ^[15].

AM Best's scale has seven rating categories, from Superior to Poor. A and A- make up the Excellent category, the second-highest, which AM Best assigns to insurers that have, in its opinion, “an excellent ability to meet their ongoing insurance obligations.” AM Best also states that a rating is not assigned to specific policies and does not address an insurer's claims-payment policies or procedures ^[15], so claims handling needs its own questions.

The broker should name the issuing insurer and its current rating, and explain who handles claims and with what experience of AI disputes. It is also worth confirming whether the policy is admitted, meaning issued by an insurer licensed in the state concerned, or written on a surplus lines basis by an insurer that is not. State guaranty funds, which protect policyholders when an insurer fails, are usually financed by assessments on the insurers each state regulates ^[17], and the protection can differ between the two.

Mayflower's coverage is written on A- (Excellent) AM Best rated paper, backed by some of the world's largest reinsurers, whose names are confidential, and the about page describes that backing. Whichever insurer a company chooses, it is sensible to agree a minimum rating with the broker before any quotes arrive.

What Will an AI Liability Underwriter Ask?

An AI liability underwriter will ask how the company governs its AI, and the answers affect both whether cover is offered and on what terms. Mayflower underwrites on the applicant's AI governance: it requires an AI governance policy and an AI system inventory among 5 required documents, and states that 8 recommended documents, such as board minutes on AI risk and bias audit results, may improve terms. It is also worth asking how renewal works when the company's AI use changes, since a new AI agent or hiring tool can alter the exposure within a year.

The recommended preparation is to assemble the governance documents before the application goes in, and the guide to how underwriters assess AI risk explains each section of the application.

What Should You Ask Your Existing Insurers?

The existing D&O, EPL and E&O insurers should be asked two things before any AI policy is bought: whether they intend to add AI exclusions at renewal, and whether their current wording responds to the company's specific AI exposures.

The first question matters because AI exclusions are spreading, with three ISO generative AI exclusions for general liability taking effect in January 2026 ^[18]. Hunton Andrews Kurth reported in 2025 that one insurer had introduced an exclusion, intended for its D&O, E&O and fiduciary liability products, that removes cover for any claim “based upon, arising out of, or attributable to” the “actual or alleged use, deployment, or development of Artificial Intelligence by any person or entity” ^[19]. The second matters because policies drafted before generative AI came into everyday use may say nothing about it, which leaves the answer to an AI claim to be argued after the loss.

A general assurance from an insurer is of limited value, as the Lloyd's Market Association noted in reporting on its 2025 survey of AI loss scenarios: “an authoritative view on coverage or exclusion of losses arising from a specific loss would require a review of the facts and circumstances and application of the policy wording(s) used” ^[20].

The request to each insurer should therefore describe concrete scenarios, such as a discrimination claim over a vendor's screening tool or a securities suit over the company's statements about its AI, and ask for the answer in writing. Where an insurer adds an exclusion or will not confirm cover, the guide to silent AI and the new AI exclusions compares the ways to fill the gap.

What Should an AI Insurance Checklist Include?

An AI insurance checklist should compare each quote on what it covers, how its limits and reporting rules work, and who issues it and handles claims. The 11 criteria below work best copied into a spreadsheet with one column per quote and completed from the policy wording rather than the quote summary.

A quote that leaves any row unanswered should go back to the broker before it is compared with the others on price.

What Should You Do Next?

The recommended course of action is to work through four steps, starting about 90 days before the main management liability renewal so that there is time to negotiate with existing insurers and to arrange affirmative cover if they will not move.

1. Prepare the claimant map and the AI system inventory.
2. Send the scope and structure questions to the broker, with a request for specimen wordings.
3. Ask the existing insurers for their position on AI in writing.
4. Compare the quotes on the checklist above, row by row.

Mayflower places its coverage through brokers, so a company without one should appoint a broker experienced in management and professional liability. The claimant map and the existing insurers' answers also suggest where to start: AI-EPL for a company that uses AI in hiring, AI-E&O for one whose AI serves clients, AI-D&O for one whose board oversees disclosures about AI, and AI DIC Excess for one whose existing insurers have added AI exclusions. Questions about how Mayflower's wordings are written can go to the team through Talk to us, and a broker ready to apply can start a quote; either way, how a policy responds to a particular claim depends on its wording, so each answer in this guide should be checked against the specimen form before the policy is bound.