AI liability insurance: what it covers and who sells it

What is AI liability insurance?

AI liability insurance is cover for claims that arise from what an AI system says, decides or does. It is built around the behavior of the model, not only around a hack or a human slip, and it is sold in three main forms.

• Standalone AI policies, such as Armilla's, which is designed to respond ahead of conventional cyber or E&O wording¹.
• Endorsements that write AI into a cyber, tech E&O or professional liability policy, such as the AI wording Beazley added in September 2026².
• Programs that combine lines, such as Vanguard AI, which pairs Chaucer's cyber and tech E&O with Armilla's AI policy under agreed rules for splitting mixed losses³.

Our register holds 28 listings. This is how they break down:

• By type: standalone policies 8, endorsements 13, programs 4, performance warranties 1, certification-backed covers 1 and planned launches 1.
• By status: confirmed on the issuer's own pages 25, trade press only 1, and announced but not yet sold 2.

Two terms come up often. Affirmative cover means the wording says in plain terms that AI losses are in scope. Silent cover means the wording says nothing either way, which leaves the answer to a claims dispute.

What does AI liability insurance cover?

It covers claims that come from an AI system's behavior. Wording differs by insurer, but these are the risks the policies we track name most often:

• Model errors, including hallucinations, inaccuracy, drift and underperformance^{1, 4}.
• Harmful or defamatory output, and IP claims over AI-generated content^{1, 5}.
• AI agent failures, such as wrong decisions, improper tool use or escalation errors¹.
• Discrimination claims over biased outputs^{5, 6}.
• Regulatory defense costs and insurable fines tied to AI use^{1, 7}.
• Privacy claims and data leakage that do not involve a breach¹.

Some products pay your own losses rather than a third party's. Relm's RESCAAI pays response and loss costs when an AI system you rely on goes wrong, including downtime, reputational damage and recalls⁵. Mosaic Insurance, working with Munich Re's aiSure team, covers financial losses when a model misses defined performance thresholds⁸.

Beazley's AI Voluntary Shutdown Endorsement is for a business that suspends its own malfunctioning AI system to prevent further financial harm⁹. Two risks have their own guides: AI hallucination insurance and insurance for AI agents.

Why might your general liability or E&O not cover AI?

Most business policies were written before generative AI and say nothing about it. That silence is now being settled in both directions: some insurers are writing AI out of their policies and others are writing it in.

The exclusions are spreading. A Trades Coverage review of admitted-market filings through 31 July 2026, reported by Beinsure, counted 4,078 filed adoptions of generative AI exclusion forms across 49 states and DC, 2,369 of them in force^{13, 14}. Filings show adoptions, not policies sold: they do not show whether an exclusion was attached to an issued policy, and the review does not capture the non-admitted market¹³.

In November 2025 the Financial Times reported that insurers had asked US regulators to approve AI exclusions. Beinsure's account names AIG, Great American and W. R. Berkley, while TechCrunch's names Chubb in place of AIG^{15, 16}. AIG has said it has no current plans to use the exclusions it filed¹⁶.

Insurance Business later reported that Chubb, Travelers, W. R. Berkley, Berkshire Hathaway and AIG had each filed to adopt ISO's forms or their own versions by April 2026, without giving states or filing numbers¹⁷. Beinsure reported that Chubb covers some AI incidents but excludes those that could cause damage on a wide scale at the same time, without naming a product or policy line¹⁵.

If you cannot tell which side your policy is on, our coverage checker shows where to look, and the exclusions tracker lists the forms and filings we follow. For E&O in detail, see does tech E&O cover AI.

Who sells AI liability insurance?

The standalone policies and programs below are written for AI risk. Larger carriers such as Beazley, AXA XL and Hiscox have mainly added AI wording to policies they already sell^{2, 18, 19}.

• AArmillaArmilla Insured$25Mup to, per organizationSurplus lines brokers, or Armilla underwriting after an AI risk assessment
• TTestudoGenerative AI Liability Insurance$10Mlimits from $1M to $10MThrough brokers, on a US excess and surplus lines basis
• RRelm InsuranceNOVAAILimit not publishedLicensed brokers
• RRelm InsurancePONTAAILimit not publishedLicensed brokers
• RRelm InsuranceRESCAAILimit not publishedLicensed brokers
• MMosaic Insurance and Munich ReAI performance cover15Minitial capacity, EUR, USD or CADThrough brokers
• CCowbell and ChaucerPrime One$10Mmaximum limitHow to buy not stated
• AAIUC and BeazleySpecialist AI liability product on Beazley paper (trade press)Limit not publishedHow to buy not stated
• CChaucer and ArmillaVanguard AI$25M+AI aggregate, alongside $10M cyberSurplus lines brokers
• MMayflower Specialty and Hadron InsuranceAI liability programLimit not publishedDirect, through a broker, or embedded through partners
• MMunich ReaiSureLimit not publishedMunich Re directly, or partner programs
• GGoogle CloudRisk Protection Program with affirmative AI coverLimit not publishedThrough a licensed broker after a free security scan

Armilla's policy is placed on Lloyd's paper through surplus lines brokers, and each policy comes with independent certification of the AI system and risk reporting¹. Armilla launched it in April 2025, co-developed with Chaucer²⁰.

Testudo, a Lloyd's coverholder, launched standalone third-party liability cover for US companies deploying generative AI on 21 January 2026²¹. It sells only through brokers, on an excess and surplus lines basis, and the cover is not offered in some states²². Atrium and QBE joined Apollo on its capacity panel in February 2026²³.

Relm, a Bermuda insurer, launched three AI policies in January 2025: NOVAAI for companies that build AI, PONTAAI as an excess wrap over existing liability programs, and RESCAAI for first-party losses⁵. Cowbell's Prime One is a non-admitted cyber policy for US businesses with $250 million to $1 billion in revenue that names AI-related incidents as covered²⁴.

The Mayflower Specialty and Hadron Insurance program has four modules: AI D&O, AI EPL, AI E&O and an AI difference-in-conditions excess layer over an existing program. Policies are issued by Hadron Specialty Insurance Company and backed by Lloyd's market reinsurers, with Aon structuring the placement²⁵. It is for US businesses, availability of each coverage varies by state, and the companies call it the first dedicated affirmative AI liability program in the US market, which is their own claim^{25, 26}.

Some listings are not yet confirmed as on sale. HSB announced small-business AI liability cover in March 2026, to be sold through partner carriers once regulators approve it²⁷. Trade press reported in May 2026 that AIUC secured Beazley capacity for a specialist AI liability product, but the only source we could read is an aggregator write-up²⁸.

Ollive pairs runtime monitoring of AI agents with planned liability cover for AI vendors. Its own page says the insurance program is being designed, with risk to be carried by a licensed insurer it does not name²⁹. Company profiles and their dated changes are on who insures AI, and the tracker log lists market events by date.

Can you add AI cover to a policy you already have?

Sometimes. Several insurers sell endorsements, or have rewritten their wording, so that AI sits inside a cyber, tech E&O or professional liability policy.

• BBeazleyAffirmative AI wordingLimit not publishedHow to buy not stated
• BBeazleyAI Voluntary Shutdown EndorsementLimit not publishedAdded to Beazley cyber policies
• BBeazleyAI Regulatory Defence & Penalties EndorsementLimit not publishedAdded to Beazley cyber policies
• CCFCAffirmative AI wording across seven productsLimit not publishedBuilt into CFC products
• CCoalitionAffirmative AI EndorsementLimit not publishedThrough brokers
• CCoalitionDeepfake Response EndorsementLimit not publishedThrough brokers
• AAXA XLGen AI endorsement to CyberRiskConnectLimit not publishedHow to buy not stated
• HHiscoxTechnology Professional Indemnity with affirmative AI coverLimit not publishedHow to buy not stated
• CCounterpartAffirmative AI CoverageLimit not publishedThrough brokers
• CCorgiModular AI coverage on tech E&OLimit not publishedOnline, direct
• EEmbrokerArtificial Intelligence Coverage Endorsement$150Kfor algorithm removalOnline, direct
• VVouch and HiscoxAI endorsement on E&O, with CorixLimit not publishedThrough Vouch
• HHSB and Munich ReAI Liability Insurance for small businesses (announced)Limit not publishedThrough partner carriers

Embroker publishes a figure: its Artificial Intelligence Coverage Endorsement, in effect since 5 August 2025, caps algorithm-removal costs at $150,000, and Embroker says other AI claims get full policy limits⁶. The other endorsements state no AI limit on the pages we read.

Beazley's three AI endorsements sit on its Full Spectrum Cyber policy: affirmative AI wording, AI Voluntary Shutdown, and AI Regulatory Defence & Penalties⁹. Its 17 September 2026 page said the wording was offered on an admitted basis in some US states as well as surplus lines². Its 6 October page lists the three as US surplus lines only, with Canada shown as preliminary, and states no limit, sublimit or retention⁹.

CFC's wording covers Tech E&O, Professional Liability, eHealth, Intellectual Property, Management Liability, Media and Cyber Proactive Response, and addresses hallucination, AI-generated content and model drift⁴. Its Media policy and cyber changes took effect on 30 July 2026³⁰. In September 2026 CFC moved most of its financial institutions products onto its Cyber Proactive Response wording, which covers AI-related cyber exposures³¹, and confirmed that IP claims involving AI stay covered where they otherwise meet the policy's terms³².

AXA XL added a generative AI endorsement to its CyberRiskConnect policy in October 2024 for companies that build their own models¹⁸. Hiscox rewrote its UK technology professional indemnity wording so client claims from the insured's use of AI are expressly covered¹⁹. Counterpart added explicit cover for errors by AI tools to its small-business professional liability lines³³. Vouch says it can place E&O for AI companies with an AI endorsement from Corix, which it describes as a division of Hiscox³⁴.

Zurich told Business Insurance in April 2026 that some insurers have issued endorsements clarifying whether AI is meant to be covered, and that endorsements broadening cover for AI-related regulatory fines are increasingly available³⁵. That is market commentary, not a Zurich product launch.

How much does AI liability insurance cost?

No insurer we track publishes prices for AI liability cover. Premiums are set in underwriting, which for AI can start with a review of the system itself.

Armilla's policies come with independent AI system certification¹, AIUC ties insurance terms to an audit against its AIUC-1 standard³⁶, and Mayflower says it scores risks such as bias, drift and hallucination²⁶. 6 of the 28 listings in our register publish a limit figure:

• Armilla: up to $25M per organization¹.
• Vanguard AI: AI aggregate limits of $25M or more, alongside $10M of cyber³.
• Cowbell Prime One: up to $10M²⁴.
• Testudo: limits from $1M to $10M, per its launch post²¹. Atrium's February 2026 release cited up to $9.25M per insured²³.
• Mosaic and Munich Re: up to 15 million of initial capacity, in euros, US dollars or Canadian dollars⁸.
• Embroker: $150,000 for algorithm removal⁶.

Sublimits can matter more than headline limits. Beinsure, relaying Financial Times reporting, said QBE offers capped cover for EU AI Act fines by endorsement, limited to 2.5% of the insured limit. We have not seen QBE's wording, the policy line or the date¹⁵.

For scale, Deloitte estimated in May 2024 that annual global AI insurance premiums could reach about $4.7 billion by 2032³⁷. That is a projection, not a measure of today's market.

How do you buy AI liability insurance?

Brokers are the most common route named in our register, including surplus lines brokers for Armilla and Vanguard AI. Corgi and Embroker sell online and direct, and Mayflower sells direct, through brokers or embedded through partners^{38, 6, 25}.

Surplus lines, also called non-admitted, cover is written by an insurer that is not licensed in your state and placed through a broker who is licensed to place it. It lets insurers write new kinds of risk on their own forms. Armilla, Testudo, Cowbell's Prime One and Beazley's AI endorsements are sold this way^{1, 22, 24, 9}.

Some AI cover arrives through a third party. Google Cloud's Risk Protection Program pairs a free security scan with cyber policies from partner insurers. Google announced affirmative AI cover for Google-related AI workloads as available in May 2025 but does not say which partner writes it, and the program as a whole is labelled Preview^{39, 40}. AIUC links insurance terms to an audit against its AIUC-1 standard for AI agents³⁶.

AI Insurance Tracker does not sell insurance. Ask a licensed broker to confirm the wording, limits and exclusions of any policy before you rely on it.

Building an AI product? The AI startup insurance guide covers tech E&O, cyber and D&O by stage.