Summary

For owners and managers in the US and Europe about to let an AI agent write to or act for customers: what existing insurance policies say about AI losses in 2026, which separate AI covers exist, and what to ask your broker in writing.

What you'll learn

  • Which three ISO endorsements can remove generative AI from US general liability since January 2026, and what each one removes
  • How to spot an AI exclusion on your policy's endorsement schedule
  • How many US insurer filings adopt AI exclusions, and what that count does not show
  • What professional liability, cyber and D&O policies say about AI errors
  • Which standalone AI liability covers exist, who can buy them and what they require
  • What the London market and Germany (our EU example) show, and the questions to put to your broker before go-live

Whether business insurance pays for a mistake made by an AI agent depends on the wording of each policy you hold, and in the US that wording changed in 2026. Since January 2026, Verisk's ISO program has offered insurers three optional endorsements that remove generative AI from commercial general liability, and a third-party count found 4,078 filed adoptions of AI exclusion forms across 49 states and the District of Columbia by July 31, 2026.[1][6] Professional liability, cyber and directors' and officers' (D&O) policies have no standard AI form, so the answer differs by policy: W.R. Berkley excludes AI outright in its D&O and E&O products, while London market underwriters describe many policies, especially cyber, as "silent" on AI. A handful of separate AI liability covers now exist, each with its own conditions.[5][7][12]

This page is for owners and managers in the US and Europe about to let an agent write to or act for customers. It sets out what each policy line says as of November 2026, which standalone covers exist, what the sources show for the London market and for Germany, and the questions to put to your broker in writing. It is not insurance advice: only your policy wording and your insurer can say whether a given loss is covered.

Does business insurance cover AI mistakes?

Business insurance covers an AI agent's mistake only where a policy's wording reaches that kind of loss and no exclusion removes it. As of November 2026, the sources we read show a split by policy line: US commercial general liability can carry an ISO generative-AI exclusion from January 2026, professional liability and cyber policies have no standard AI form that we found and have to be checked policy by policy, and at least one specialty carrier excludes AI outright in its D&O and E&O products.[1][4][7] The table sums up what each line says and how solid the evidence is.

Policy lineCould it respond to a loss an agent causes?AI exclusion seenSource and date
US commercial general liability (CGL)For the injury and damage types the policy lists, unless an AI endorsement is attachedThree optional ISO endorsements, edition January 2026; 4,078 filed adoptions to July 31, 2026ISO forms; Trades Coverage, Aug 2026[2][3][6]
US professional liability (E&O)Depends on each policy's wording; brokers are told to confirm whether E&O picks up the exposure or it falls between policiesW.R. Berkley: an "absolute" AI exclusion in its E&O, D&O and fiduciary products, introduced 2025Insurance Business, Sept and Oct 2026; Insurance Journal, Aug 2026[4][5][7]
US cyberDepends on each policy's wording; the same broker check as for E&ONone named in the US sources we readInsurance Business, Oct 2026[4]
US directors and officers (D&O)Expected to be one of the first lines where AI exclusions are testedW.R. Berkley's absolute AI exclusion (as above)Insurance Journal, Aug 2026[7]
London market, especially cyberAI mostly "silent": neither named nor excludedLimited AI-specific wording seenIUA, March 2026[12]
German liability on GDV model termsPersonal injury and property damage; pure financial loss from services or advice excluded in the model terms, and covered by classic business liability usually not or only in part (Bitkom)No AI wording in the model terms; no published AI exclusion from a German insurer foundGDV, Oct 2026; Bitkom, Oct 2026[18][19]
Standalone AI liability coverWritten for AI losses, on each product's own conditionsNot applicableVendor pages and releases, Mar–Nov 2026[9][11][14]

Two limits apply to every row. No court has yet decided how an AI exclusion applies: Alana McMullin of the law firm Lathrop GPM told Insurance Journal in August 2026 that she expects the first tests in E&O and D&O and that no bellwether decision existed yet, and we found none as of November 2026.[7][8] And who is legally liable for an agent's mistake is a separate question, set out in who is liable for an AI agent's mistakes in Europe. This page covers what insurance may pay once a claim exists.

What is a generative AI exclusion?

A generative AI exclusion is an endorsement that removes claims arising out of generative AI from a liability policy. Since January 2026, Verisk's ISO program, which publishes standard policy forms that insurers can adopt and file, has offered three such endorsements for US commercial general liability, edition 01 26. They are optional: each insurer decides whether to attach one, so two carriers' policies can differ.[1][4]

ISO formWhat it removesPart of the policy it changes
CG 40 47 01 26Bodily injury, property damage and personal and advertising injury connected to generative AI; described in trade and legal reporting as the broadest of the three (we did not read its text)Coverage A and Coverage B of the CGL[1][5][8]
CG 40 48 01 26Personal and advertising injury arising out of generative AICoverage B only[2]
CG 35 08 01 26Bodily injury and property damage arising out of generative AIThe separate products and completed operations liability coverage part[3]

The two forms we read in full use the same wording: the insurance does not apply to the listed injury or damage "arising out of 'generative artificial intelligence'".[2][3] They define generative AI as a machine-based system or model trained on data that can create content or responses, and they give text, images, audio, video and code as examples without closing the list.[2][3] According to Verisk, as reported by Independent Agent, the definition is based on government and NAIC definitions.[1]

The trigger is the connection to AI, not how much of the work AI did: Insurance Business reported that the forms contain no proportion test.[5] Our reading: an agent built on a language model that drafts customer emails or quotes falls within the definition as written, because it creates text responses. Whether a given claim "arises out of" the agent is the kind of question that, so far, no court has answered.

What to look for on the endorsement schedule. The endorsement schedule lists every form attached to your general liability policy by number and edition. Look for CG 40 47 01 26, CG 40 48 01 26 or CG 35 08 01 26, and for any form with "artificial intelligence" in its title, since carriers also file wording of their own. Insurance Business, reporting one broker's warning in October 2026, wrote that brokers need to read the schedule at every renewal rather than assume the standard wording applies.[4]

How many insurers use AI exclusions?

No public source we found counts how many insurers put AI exclusions on policies; the best count covers filings. A third-party analysis counted 4,078 filed adoptions of AI exclusion forms under 287 carrier or program labels across 49 states and the District of Columbia by July 31, 2026, of which 2,369 had reached their effective date.[6] Trades Coverage built the count from SERFF public filings and Florida's I-File system, so it covers the admitted market only; surplus lines policies are not in it. Each record is one carrier, one form and one state, and the 287 are filing labels, not a count of insurers. A filing shows that an insurer intends to use a form, not that the form is on any particular policy.[6]

The carriers with the most filed adoptions in that analysis were:

CarrierFiled adoptions to July 31, 2026
National Union Fire Insurance Company of Pittsburgh, Pa.313
Allianz Global Risks US234
Atlantic Specialty200
Star152
XL Insurance America146

Source: Trades Coverage, updated August 18, 2026.[6]

For contractors, the trades its report covers, the same analysis did not identify an admitted standalone generative-AI product in the filings it reviewed that replaces the excluded cover.[6] Outside general liability, W.R. Berkley introduced an "absolute" AI exclusion for its D&O, E&O and fiduciary products in 2025, as reported by Insurance Journal and Insurance Business; Insurance Journal could not reach Berkley for comment.[5][7]

Some reporting describes the 2,369 effective records as exclusions "in force" on policies.[5] The underlying count measures filings, so we use the filing figure. Our reading of what the count does show: a US general liability policy renewed in 2026 can differ from the previous year's policy in the endorsement list alone, so the schedule is the place to check, not last year's answer.

Do professional indemnity and cyber policies cover AI errors?

Whether professional liability (errors and omissions, E&O, called professional indemnity in the UK) or cyber cover responds to an AI error depends on each policy's wording; no market-wide rule answers it. Insurance Business wrote in October 2026 that brokers need to confirm whether a client's E&O or cyber policy picks up an AI exposure, or whether it falls between policies, and W.R. Berkley excludes AI outright in its E&O, D&O and fiduciary products.[4][5][7] No court decision on an AI claim under either line had been reported in the sources we read, as of November 2026.[7][8]

The evidence here is thinner than for general liability, for three reasons:

  • No standard form. We found no standard AI endorsement for E&O or cyber comparable to the ISO forms, so in our reading the answer for one insurer says little about another.
  • Carriers are moving in both directions. W.R. Berkley excludes AI outright in its E&O and D&O products, while Alana McMullin of Lathrop GPM said some carriers may choose to price AI risk instead of excluding it.[7]
  • Silence cuts both ways. In our reading, a policy that does not mention AI neither promises nor refuses cover for it. A London market roundtable of the International Underwriting Association reported in March 2026 that many policies, especially cyber, treat AI as "silent" in this sense.[12]

Our reading of these sources: for an agent that sends a customer a wrong quote or wrong information, and the customer claims a financial loss, the policy most likely to be asked first is professional liability, and the answer turns on its wording rather than on any market-wide rule. That is a question for your broker, in writing, before the agent goes live.

Insurance Business wrote in October 2026 that documenting how clients use AI, and the human checks in place, is likely to become part of the underwriting conversation, and reported in September 2026, citing Aon, that carriers evaluating commercial accounts are increasingly looking for documented AI governance.[4][5] A company that can describe what its agent does, which systems it can reach and who approves its output has those answers ready.

What is AI liability insurance and who sells it?

AI liability insurance is cover written specifically for losses from a business's use of AI, rather than cover that simply does not exclude it; Armilla calls its product "affirmative" for that reason. As of November 2026 the products we could verify are HSB's AI Liability Insurance (announced March 18, 2026, pending regulatory approval), Armilla's Lloyd's-backed cover for the US and Canada, Testudo at Lloyd's, Munich Re's aiSure, and startup offers from Klaimee and Corgi.[9][11][12][14][15][16]

ProductWho writes itWho can buyWhat it responds toStatus and limits as stated
HSB AI Liability InsuranceHSB (Munich Re), added to partner carriers' policiesSmall and medium-sized companies, through carriers that partner with HSB; the release states no size cap or countryBodily injury, property damage, personal and advertising injury from the insured's use of AIPending regulatory approval at announcement; not sold directly[9][10]
Armilla Affirmative AI Liability InsuranceCertain underwriters at Lloyd's (Chaucer, AXIS, Convex); Armilla is the coverholderBusinesses in the US and Canada, through licensed brokersAI model error, output claims, AI agent mistakes, among othersUp to US$25 million aggregate per insured and US$5 million per model; not bound until confirmed in writing[11]
TestudoLloyd's coverholder, launched January 2026Enterprises deploying generative AIThird-party AI liabilityCapacity reported at up to US$9.25 million per insured[12][13]
Munich Re aiSureMunich ReAI companies and corporates adopting AIFour covers named on its FAQ page: Contractual Liabilities (performance guarantee), Own Damages, General Liability and IP LiabilityPremium set by model quality and performance stability, after an evaluation of the model development pipeline[12][14]
KlaimeeStartup; no carrier named on its pageStartups and enterprises deploying AI agentsEvaluates and scores an agent, then insures itNo carrier or limits published; risk score sets the price; underwriting a first group of customers[15]
Corgi Tech E&O & AI LiabilityCorgi Insurance Company or Technology Risk Retention GroupStartupsTechnology errors and AI liabilityA risk retention group is not subject to all state insurance laws and has no guaranty-fund protection[16]

HSB gave its reason for the product in a survey published with the release: 74% of the 1,000 "Main Street" businesses it surveyed use AI and 91% plan to.[9] HSB's product page names no exclusions, limits or states.[10] We found no later source confirming regulatory approval or naming a partner carrier, as of November 2026.

Which AI liability covers fit an established company?

Our reading of who each product is built for: HSB's endorsement is designed for small and medium businesses buying through an existing carrier, so it is the one to ask your broker about by name. Armilla sells through licensed brokers and Testudo writes for enterprises; their published limits run to millions of dollars per insured, which suggests larger buyers.[11][12][13] Munich Re's aiSure starts from an evaluation of the model development pipeline, which the International Underwriting Association calls "invasive" for performance guarantees; for a company whose agent is built by a vendor, that is a conversation the vendor would have to join.[12][14] Corgi's page addresses startups, and Klaimee's names startups and enterprises deploying AI agents.[15][16]

None of these products publishes a price in the sources we read. Klaimee says its risk score sets the insurance price, and Munich Re says model quality and performance stability determine the aiSure premium.[14][15] In our reading the others are also priced case by case, but no source says so.

Is AI insurance different in the London market and in Germany?

Yes, in the sources we read: the ISO endorsements are US forms, London market underwriters describe AI as mostly "silent" in existing policies, and the German model liability terms do not mention AI. The International Underwriting Association's clauses committee reported in March 2026 that it had seen limited AI-specific wording in the London market so far, and that the definition of "Computer System" in London cyber clauses may be broad enough to include AI as software.[12] We checked Germany as the EU example; we did not check other EU countries.

In Germany, the starting point is the insurers' association GDV's model terms for business and professional liability (AVB BHV, October 2026 version, non-binding). They cover legal liability for personal injury, property damage and the financial loss that follows from them, but pure financial loss only in a listed form that excludes claims from work or services the insured delivers and from advising, planning, testing or expert activity.[18] The model terms do not mention AI.[18]

Bitkom, the German digital industry association, assessed the German market in October 2026 for insurers and brokers. Each point has its own scope:[19]

  • Liability. Classic business liability covers pure financial loss usually not at all or only in part, unlike special concepts such as cover for IT service providers. Bitkom does not rule out that insurers will propose an AI exclusion.
  • Property. In all-risk property insurance, a general exclusion of property damage caused by AI is not currently market standard; Bitkom calls this state "Silent AI".
  • The company's own losses. Pure financial losses a company itself suffers from an AI malfunction are not yet widely offered as standalone cover and mostly stay with the company, unless they trace back to an established insured event such as a cyberattack.
  • AI manufacturers. Their risks are generally insured through technology E&O.

Aon, writing about large industrial clients in September 2026, assessed that AI damage is silently co-insured in many liability contracts and expects insurers to reshape their liability terms.[17] Our reading for a German company: if an agent sends a wrong quote and the customer's loss is purely financial, the answer turns on whether the company holds financial-loss liability cover or a business liability policy extended for it.

The EU Product Liability Directive treats software as a product, and Germany must transpose it by December 9, 2026.[17] It governs who can claim, which is covered in who is liable for an AI agent's mistakes in Europe.

What questions should you ask your broker about AI?

Ask your broker in writing, before the agent goes live and again at each renewal, which of your policies would respond to a loss the agent causes and which AI exclusions are attached. Insurance Business wrote in October 2026 that brokers need to read the endorsement schedule at every renewal and that documenting AI use and human checks is likely to become part of the underwriting conversation.[4] The questions below are our list, built from the forms and sources on this page.

  1. Which endorsements are attached to our general liability policy? Ask specifically whether CG 40 47 01 26, CG 40 48 01 26, CG 35 08 01 26 or a carrier's own AI exclusion is among them.
  2. Which policy would respond if our agent gives a customer wrong information? Describe a concrete case, such as a wrong quote that costs the customer money, and ask which policy you would claim under and why.
  3. Do our E&O, cyber and D&O policies mention AI anywhere? Ask about exclusions and about definitions, such as "computer system", that could reach AI software.
  4. Will an AI exclusion be added at renewal? Ask for the renewal endorsement list in advance, not after binding.
  5. What will the underwriter ask about our AI use? Ask which documents they want: what the agent does, what it can reach, who approves its output.
  6. Is affirmative AI cover available to us, and from which carrier? Name HSB's AI Liability Insurance and ask about status, carrier, trigger and limits for any product offered.
  7. Does our cover change if a vendor builds or runs the agent? Ask how the policy treats output from software a third party built and operates for you.
  8. If we trade in Germany: is pure financial loss from our services covered? Ask whether the business liability policy was extended for it, or whether separate financial-loss cover is needed.
  9. Will you confirm the answers in writing? An email answer to each question is a record you can return to when a claim arises.

Download the broker questions checklist (PDF). It carries the same nine questions, the three ISO form numbers and the date the facts were checked, ready to forward to your broker.

Which AI controls can you show a broker or insurer?

For a company that buys no standalone AI policy, the controls around the agent are what a broker or underwriter can be shown: approval before the agent sends anything, an audit log of what it did, and access scoped to the systems it needs. That is our inference from the broker advice above, not an underwriting rule. How far an agent can reach and running AI agents after launch set out those controls; what to do when your AI agent makes a mistake covers the hours after an incident, when the broker's written answers become useful.

The forms, the filing counts and the product list on this page will change. The one question that stays the same is the first on the list: which forms are on our policy, and what do they say about AI.

Frequently asked questions

How much does AI liability insurance cost? None of the free sources we read publishes a price for AI liability insurance, as of November 2026. Armilla sells through licensed brokers and binds cover only once it is confirmed in writing, and HSB's product is added to partner carriers' policies, which suggests the premium is set per application. Estimates circulating on vendor and broker blogs could not be verified, so we quote none.

Does liability insurance cover AI? It depends on the policy's wording. Since January 2026, insurers using Verisk's ISO forms can attach three optional endorsements (CG 40 47, CG 40 48 and CG 35 08) that remove losses arising out of generative AI from US general liability. A third-party count found 4,078 filed adoptions of AI exclusion forms across 49 states and DC by July 31, 2026. Filings show intent to use a form, not that it is on your policy, so check your endorsement schedule.

Does cyber insurance cover AI mistakes? There is no standard answer: we found no AI form for cyber policies comparable to the ISO endorsements, so it depends on each policy's wording. Insurance Business wrote in October 2026 that brokers need to confirm whether a client's cyber or E&O policy picks up an AI exposure, or whether it falls between policies. In the London market, an International Underwriting Association roundtable reported in March 2026 that many policies, especially cyber, treat AI as “silent” and that the definition of a computer system in London cyber clauses may be broad enough to include AI as software. No court has tested either reading that we found.

Is HSB AI liability insurance available? HSB, part of Munich Re, announced AI Liability Insurance for small and medium businesses on March 18, 2026. It covers bodily injury, property damage and personal and advertising injury from a business's use of AI, is added to partner carriers' policies rather than sold directly, and was pending regulatory approval at announcement. We found no later source confirming approval or naming a partner carrier, as of November 2026.

Notes

  1. Independent Insurance Agents & Brokers of America, Independent Agent, "Verisk to Roll Out New General Liability Exclusions for Generative AI Exposures", October 21, 2025 (reporting Verisk's announcement; Verisk's own pages did not load). independentagent.com. Checked November 19, 2026.
  2. Insurance Services Office, Inc., CG 40 48 01 26, "Exclusion – Generative Artificial Intelligence (Coverage B Only)", © 2025, copy hosted by ALM. assets.alm.com (PDF). Checked November 19, 2026.
  3. Insurance Services Office, Inc., CG 35 08 01 26, "Exclusion – Generative Artificial Intelligence" (Products/Completed Operations Liability Coverage Part), © 2025, copy hosted by ALM. assets.alm.com (PDF). Checked November 19, 2026.
  4. Insurance Business, "AI exclusions are spreading through commercial policies, brokers warn clients to check cover", October 21, 2026 (reporting the warning of broker Joe Gendelman, Bruce Gendelman Insurance). insurancebusinessmag.com. Checked November 19, 2026.
  5. Insurance Business, "ISO's generative AI exclusion is already on thousands of CGL policies", September 16, 2026. insurancebusinessmag.com. Checked November 19, 2026.
  6. Trades Coverage, "GenAI exclusion adoption" report, updated August 18, 2026; analysis of SERFF public filings and Florida I-File, admitted market only, data to July 31, 2026; counts are carrier × form × state. tradescoverage.com. Checked November 19, 2026.
  7. Insurance Journal, magazine feature, August 17, 2026 (quoting Alana McMullin, Lathrop GPM). insurancejournal.com. Checked November 19, 2026.
  8. Shumaker, "Client Alert: The New AI Coverage Fight: Exclusions, Endorsements and Denied Claims", National Law Review, July 30, 2026. natlawreview.com. Checked November 19, 2026.
  9. HSB (Munich Re), "Introducing AI liability insurance for small businesses", press release, March 18, 2026, including survey footnote (micro 1–10, small 11–50, medium 51–500 employees; these define the survey sample, not eligibility). munichre.com. Checked November 19, 2026.
  10. HSB, "AI Liability Insurance", product page. munichre.com. Checked November 19, 2026.
  11. Armilla, "Affirmative AI Liability Insurance", company website (no policy date shown). armilla.ai. Checked November 19, 2026.
  12. International Underwriting Association, IUA Circular 033/2026, "Clauses Committee Roundtable: AI Insurance", March 2026 (summary of a roundtable discussion). iua.co.uk (PDF). Checked November 19, 2026.
  13. FinTech Global, "Testudo expands AI liability capacity to $9.25m", March 9, 2026 (states limits of up to $9.25m per insured). fintech.global. Checked November 19, 2026.
  14. Munich Re, "aiSure: FAQ", undated page. munichre.com. Checked November 19, 2026.
  15. Y Combinator, "Klaimee insures your AI agents", launch page, undated. ycombinator.com. Checked November 19, 2026.
  16. Corgi, "Tech E&O & AI Liability", company website. corgi.insure. Checked November 19, 2026.
  17. experten.de, article on Aon's assessment of the industrial liability market and the EU Product Liability Directive, September 9, 2026. experten.de. Checked November 19, 2026.
  18. Gesamtverband der Deutschen Versicherungswirtschaft (GDV), "Allgemeine Versicherungsbedingungen für die Betriebs- und Berufshaftpflichtversicherung (AVB BHV)", non-binding model terms, Stand: Oktober 2026, sections A1-3.1 and A1-6.12.1. gdv.de (PDF). Checked November 19, 2026.
  19. Bitkom, "Verantwortungsvoller KI-Einsatz: Risiken verstehen, steuern und versichern", guide, October 2026, chapters 5, 6 and 9. bitkom.org (PDF). Checked November 19, 2026.