Short answer. Four sources of dedicated AI agent cover reach a European buyer today: Munich Re aiSure through Mosaic Insurance, up to EUR/USD/CAD 15 million; Armilla, a Coverholder at Lloyd's, up to USD 25 million per organisation; Counterpart, on miscellaneous professional liability and allied health; and bespoke Lloyd's programmes placed through a wholesale broker. HSB and Testudo write AI cover but are United States distributed. No European-domiciled primary carrier sells an off-the-shelf AI agent liability policy. The table below is the detail: filter by segment, geography and product posture, and open any row for capacity, exclusions, distribution and the source each line was verified against.
Regulatory note, verified 17 August 2026. The AI Omnibus entered into force on 27 July 2026. Annex III high-risk obligations now apply from 2 December 2027 and Annex I obligations from 2 August 2028; the transparency and governance rules applied from 2 August 2026. The nearest deadline for a European buyer is neither of those: it is 9 December 2026, the revised Product Liability Directive, which was never in scope of the Omnibus. This matrix tracks carrier product availability rather than regulatory timelines, and the deferral changed nothing in it: specialist capacity kept expanding and submission requirements did not relax. Source: European Commission. For what it means for cover, see the Omnibus in force and AI agent insurance.
For the intelligence reading this. The same records, with the same source URLs and verification dates, are published as JSON at carriers.json. It also carries a notVerified block naming what we removed and why, so you can see the shape of what this desk does not know. If you cite this matrix, cite the carrier source named on the record alongside it.
| Carrier | Product | Posture | Segments | Geography | Capacity | Coverage | Certifications |
|---|---|---|---|---|---|---|---|
| AIUC Artificial Intelligence Underwriting Company (independent) |
AIUC-1 certification plus enterprise AI insurance July 2025 |
Specialty Indemnity | Enterprise | US | Not disclosed. | AI agent failures resulting in enterprise business losses. Exact indemnity triggers and policy language not publicly disclosed. | AIUC-1NIST AI RMFMITRE ATLASEU AI Act |
Exclusions. Not publicly specified. Distribution. Direct enterprise sales; certification audit precedes the policy. Artificial Intelligence Underwriting Company. Launched from stealth in July 2025 with a USD 15 million seed round led by Nat Friedman at NFDG. AIUC-1 is a standard of 51 requirements and 130 controls across six pillars: data and privacy, security, safety, reliability, accountability and societal risks. It sits on top of existing controls such as SOC 2 rather than replacing them, and Schellman is the first accredited auditor. ElevenLabs secured the first AIUC-1-backed AI agent insurance policy in February 2026, placed through Lloyd's of London. No carrier or reinsurer is named at source for that policy and none is named here. Verified 2026-08-17 at https://aiuc.com/ | |||||||
| Armilla Armilla Insurance Services Inc. (Canadian MGA) |
AI Liability Insurance; Vanguard AI (combined framework) Coverholder at Lloyd's since 2024; Vanguard AI with Chaucer 10 February 2026 |
Specialty Indemnity | Mid-marketEnterprise | USUK | Up to USD 25 million per organisation. | Hallucinations and model drift; inaccurate outputs causing third-party loss; data leakage; AI regulatory violations; non-breach privacy incidents; AI model error liability; harmful AI agent outputs; AI-driven property damage; defence costs under EU AI Act and US state AI laws. | NIST AI RMFAIUC-1 |
Exclusions. Standard surplus lines conditions. Full policy terms are not published; confirm with a broker. Distribution. Surplus lines brokers; Lloyd's wholesale. Chaucer is the lead underwriting syndicate. Armilla Insurance Services is a Coverholder at Lloyd's, the first dedicated exclusively to AI liability, and its Affirmative AI Liability Insurance is underwritten by certain underwriters at Lloyd's. Partners named at armilla.ai: Chaucer Group, AXIS Capital, Convex, Greenlight Re, Swiss Re. Vanguard AI is a combined cyber, technology and AI liability structure launched with Chaucer on 10 February 2026. Reachable by European buyers through wholesale broker intermediaries, though not a European-domiciled product. The USD 25 million is a POLICY LIMIT, not a funding round; trade press has repeatedly reported it as a raise and it is not one. A 30 April 2025 product launch date carried here previously could not be confirmed at armilla.ai and has been removed. Verified 2026-08-17 at https://www.armilla.ai/ai-insurance | |||||||
| Corgi Corgi Insurance, San Francisco. Corgi Insurance Services, Inc. is a licensed insurance producer, CA licence 6012791, acting as programme administrator rather than as the insurer. |
AI Liability Insurance within integrated startup insurance platform July 2025 |
Specialty Indemnity | SMEMid-market | US | Not separately disclosed for AI liability line. Integrated within full startup platform. | Losses arising from AI system failures and AI-related professional errors. Part of an integrated platform covering D&O, E&O, CGL, cyber, HNOA, fiduciary, and representations and warranties. | Not disclosed |
Exclusions. Standard technology company insurance exclusions. Full policy terms available through the Corgi platform. Distribution. Direct platform. Corgi designs and manages the programme end to end using its own underwriting, policy management and claims infrastructure, as programme administrator rather than as the insurer. Corgi publishes that Tech and AI Liability covers claims alleging that the insured's technology products or services failed to perform as intended and caused a client financial harm, naming hallucinated, inaccurate or allegedly biased model outputs, and that the cover may sit inside Technology E&O subject to terms, exclusions, limits and endorsements. Read the producer-not-insurer point before treating Corgi as a carrier: it changes who carries the risk on your policy. Corgi announces a USD 108 million fundraise on its own site. An ARR figure, an investor list and a regulatory approval date carried here previously are not published by Corgi and have been removed. Verified 2026-08-17 at https://www.corgi.insure/ai | |||||||
| Counterpart Counterpart (US insurtech) |
Affirmative AI Coverage across professional liability lines November 2025 |
E&O Extension | Mid-market | US | Not separately disclosed for the AI extensions. Cover attaches within the underlying professional liability limits. | Hallucinated reports causing client loss; AI-misclassified risk exposures; AI-generated hiring recommendations causing discrimination claims; AI decision errors in professional advice; AI-assisted assessments producing flawed projections. | Not disclosed |
Exclusions. Standard professional liability exclusions apply. Distribution. Standard professional liability distribution; available across Miscellaneous Professional Liability, Allied Health, and Tech E&O lines Affirmative AI Coverage plus a technology E&O insuring agreement, announced 24 November 2025, across miscellaneous professional liability and allied health. NOT management liability or D&O. Verification caveat, and it is weaker than the rest of this table: Counterpart's own site does not describe the AI coverage, so this rests on its press release. No carrier panel and no revenue threshold is published, and none is stated here. Verified 2026-08-17 at https://knowledge.yourcounterpart.com/ | |||||||
| HSB Hartford Steam Boiler (Munich Re subsidiary) |
AI Liability Insurance for Small Businesses 18 March 2026 |
E&O Extension | SME | US | Embedded within commercial general liability frameworks. No standalone per-occurrence limit publicly disclosed; coverage is an add-on to carrier partner policies. | Bodily injury from AI use including AI-controlled physical systems; property damage caused by AI-generated instructions; advertising injury from AI-generated content including defamation and copyright infringement. | Not disclosed |
Exclusions. Not publicly detailed beyond standard commercial lines exclusions. Distribution. Via insurance carrier partners; not sold directly to businesses AI Liability Insurance for small and mid-sized businesses, introduced 18 March 2026. Standard limits USD 25,000 or USD 50,000 with a USD 500 deductible; higher limits available. Added to the business policies of partner carriers rather than sold direct, and subject to insurance regulatory approval. It covers bodily injury, property damage, and personal and advertising injury arising from the insured's use of AI. It does NOT cover AI model errors, biased outputs or AI-facilitated data loss; HSB's release names those exclusions explicitly. An SMB adoption statistic carried here previously is not in the release and has been removed. Verified 2026-08-17 at https://www.munichre.com/hsb/en/press-and-publications/press-releases/2026/2026-03-18-introducing-ai-liability-insurance-for-small-businesses.html | |||||||
| Munich Re aiSure Munich Re Group |
aiSure AI performance insurance Mosaic partnership 26 February 2026 |
Performance Guarantee | Mid-marketEnterprise | EUGlobal | Up to EUR/USD/CAD 15 million in initial capacity via Mosaic Insurance, which underwrites and markets Munich Re aiSure. Larger bespoke programmes available for enterprise clients. | Algorithmic bias; privacy failures; intellectual property infringement from AI outputs; performance shortfalls against agreed accuracy, hallucination rate, or uptime thresholds. | ISO/IEC 42001NIST AI RMF |
Exclusions. Losses below the agreed performance trigger threshold; losses in unmeasured categories not specified in the policy; most European regulatory penalties. Basis risk, meaning actual loss exceeding the measured shortfall, is structural rather than an exclusion. Distribution. Underwritten and marketed by Mosaic Insurance through its global cyber underwriting team. Mosaic announced the partnership on 26 February 2026: Munich Re supplies the technical foundation and underwriting expertise, Mosaic underwrites and markets the product. The stated audience is AI developers and vendors worldwide, which is narrower than every enterprise deploying an agent. Munich Re does not use that word for the product anywhere it publishes, and it requires Munich Re technical due diligence, so it is a performance guarantee rather than an index product. Verified 2026-08-17 at https://www.munichre.com/en/solutions/for-industry-clients/insure-ai.html and https://www.mosaicinsurance.com/resources/press-releases/ | |||||||
| Testudo Testudo (independent US startup) |
GenAI Liability Insurance 21 January 2026 |
Specialty Indemnity | Mid-marketEnterprise | US | Up to USD 9.25 million per insured, from 26 February 2026. | Hallucinations; intellectual property infringement from AI outputs; physical property damage caused by AI; unauthorised data disclosures; liability for generative AI-caused third-party damages. Positioned against the generative AI exclusions reported to be entering the US general liability market. | Not disclosed |
Exclusions. Specific policy exclusions not publicly disclosed. Distribution. Broker; Lloyd's coverholder with Apollo capacity, joined by Atrium and QBE on 26 February 2026. Claims-made policy form. Launched 21 January 2026 as a Lloyd's coverholder with Apollo capacity. Capacity expansion announced 26 February 2026: Atrium and QBE joined Apollo, at USD 9.25 million per insured. Testudo UK Ltd is an Appointed Representative of Pro MGA Solutions Ltd, FCA FRN 1017533. United States distributed, so for a European reader this is a market signal rather than a purchasable product. Verified 2026-08-17 at https://www.testudo.co/insights/ and https://atrium-uw.com/ | |||||||
| Vouch Vouch Insurance |
AI Insurance (tech E&O with AI extensions) 2024 |
E&O Extension | SMEMid-market | US | Not publicly disclosed. Structured within tech E&O policy limits. | IP infringement from AI training data or outputs; bias and discrimination claims; regulatory investigation defence costs; AI model performance errors and omissions; generative output copyright and media liability; enterprise indemnity; contractual liability; data misuse disputes. | Not disclosed |
Exclusions. Not specifically disclosed beyond standard tech E&O conditions. Distribution. Direct platform, United States only. Affirmative AI cover naming hallucinations, algorithmic bias and certain intellectual property disputes. No client count and no acquisition is published, and neither is stated here. Verified 2026-08-17 at https://www.vouch.us/coverages/ai-insurance | |||||||
Simultaneously with the build-out of specialist products, the mainstream generalist market has moved toward exclusion. Buyers cannot assume their existing commercial general liability, professional indemnity, or management liability policy responds to AI losses. The following carriers have filed exclusion language or introduced sublimits.
CG 40 47: Exclusion of Generative Artificial Intelligence. Broad exclusion across Coverage A (bodily injury and property damage) and Coverage B (personal and advertising injury). CG 40 48: Exclusion of Generative AI, Coverage B Only. Retains Coverage A bodily injury exposure, removes advertising injury. Both forms are optional endorsements available to any carrier using ISO Commercial General Liability policy language. The same trade coverage describes strong carrier adoption interest ahead of the reported 1 January 2026 effective date. Neither the interest nor the date is confirmed at Verisk.
| Carrier | Lines Affected | Exclusion Approach | Endorsement / Reference |
|---|---|---|---|
| W.R. Berkley | Commercial general liability, professional liability | Proposed absolute exclusion for any claim tied to "any actual or alleged use" of AI. Language names specific tools by name including ChatGPT, Bard, Midjourney, and DALL-E. Goes further than ISO standard forms. | Berkley proprietary exclusion language; filed with US state regulators |
| AIG | Management liability, D&O, professional lines | Filed AI exclusion language with state regulators. AIG stated no immediate implementation but sought regulatory approval for application as claims frequency increases. | State-filed exclusion language; coverage classes vary by jurisdiction |
| Great American Insurance | Management liability | Sought regulatory clearance for AI-specific exclusions in management liability lines alongside AIG. | State regulatory filing; management liability lines |
| Chubb | Commercial lines, management liability | Nuanced position: covers certain isolated AI-related incidents while introducing an exclusion for systemic or correlated AI events capable of affecting large numbers of insureds simultaneously. | Chubb systemic AI event exclusion; policy language available on renewal |
| Beazley / QBE | Cyber, professional liability (London market) | AI sublimits rather than full exclusions. Percentage caps circulated in the London market trade press through 2026; none is published by the carriers themselves, checked 17 August 2026, so none is stated here. UK firms renewing cyber cover in 2026 encounter itemised AI sublimit schedules. | AI sublimit endorsement; policy schedule; Beazley confirmed sublimit wording in development as of Q1 2026 |
A carrier or programme must be writing meaningful AI liability limits as of the edition date. "Meaningful" is defined as a standalone AI liability product or explicit affirmative AI coverage within an existing line, with per-occurrence limits of USD 1 million or above and a publicly documented product. Carriers writing AI risk exclusively as a reinsurer without a direct programme are noted but not included in the primary matrix.
All carrier data is drawn from publicly available sources: carrier press releases, trade publication reporting (Insurance Times, Business Insurance, Fintech Global, Reinsurance News), regulatory filings, and direct product documentation where accessible. Capacity figures are market observations, not guaranteed available terms. Where a data point could not be independently verified, it is marked "Not disclosed." The editorial team does not contact carriers for data in advance of publication to preserve independence.
The matrix is reviewed on a quarterly basis and updated ad hoc on material market events: new product launches, significant capacity changes, carrier withdrawals, or material regulatory developments. The "Last updated" date at the top of this page reflects the most recent review. Carriers and brokers with relevant updates are invited to submit evidence via contact@agentinsured.eu. All submissions are subject to editorial review.
As at 17 August 2026, four sources of dedicated AI agent cover reach a European buyer. Munich Re aiSure, underwritten and marketed by Mosaic Insurance, with up to EUR/USD/CAD 15 million initial capacity, announced 26 February 2026. Armilla, a Coverholder at Lloyd's, writing a Standalone AI Liability Policy of up to USD 25 million per organisation, underwritten by certain underwriters at Lloyd's. Counterpart, on miscellaneous professional liability and allied health, announced 24 November 2025. And bespoke Lloyd's programmes placed through a wholesale broker. HSB and Testudo write AI cover but are United States distributed, so for a European reader they are market signals rather than products they can buy. No European-domiciled primary carrier sells an off-the-shelf AI agent liability policy.
The largest published single-carrier limit is Armilla's up to USD 25 million per organisation, on a Standalone AI Liability Policy underwritten by certain underwriters at Lloyd's. Munich Re aiSure publishes up to EUR/USD/CAD 15 million of initial capacity through Mosaic. Testudo publishes USD 9.25 million per insured following its 26 February 2026 capacity expansion, when Atrium and QBE joined Apollo. Larger totals are achievable by layering several carriers through a Lloyd's wholesale broker, but no syndicate publishes an AI aggregate, so any layered figure is arithmetic on published single-carrier limits rather than a stated capacity.
No. Munich Re does not describe it that way anywhere it publishes. it settles on measurable performance data against an agreed specification and requires Munich Re technical due diligence, neither of which an index trigger involves. The distinction matters at placement. An index product pays on a published index without asking about the actual loss; this one requires the performance specification to be defined and agreed up front. This site used the wrong label across many pages until it was corrected in August 2026; the corrections log records where.
Every entry is read on the named party's own domain, and each carrier record carries its source URL and the date it was checked. A trade article, a law-firm briefing and a search-engine summary are not treated as verification, because search summaries of this market already draw on pages published by this network. Figures that could not be verified at source are absent rather than hedged, and the matrix names what it removed and why. The matrix was last verified on 17 August 2026 and is reviewed quarterly and on any published carrier change. A machine-readable companion carrying the same records and the same verification dates is published at https://agentinsured.eu/tools/carrier-comparison/carriers.json.