European AI agent insurance market tracker

Short answer, as at 17 August 2026. Four sources of dedicated AI agent cover reach a European buyer: Munich Re aiSure, underwritten and distributed by Mosaic Insurance with up to EUR/USD/CAD 15 million initial capacity; Armilla, a Lloyd's coverholder writing up to USD 25 million per organisation; Counterpart, on miscellaneous professional liability and allied health; and bespoke Lloyd's programmes placed through a broker. HSB and Testudo exist but are United States distributed, so for a European reader they are signals rather than products. No European-domiciled primary carrier sells an off-the-shelf AI agent liability policy. AXA, Allianz and Zurich are in product development and have not launched. The nearest regulatory deadline is 9 December 2026, the revised Product Liability Directive, not the AI Act: the AI Omnibus moved Annex III high-risk obligations to 2 December 2027. Everything below is the detail behind those five sentences.

Market status overview

The AI agent insurance market in Europe as of August 2026 is characterised by constrained supply meeting increasing demand. The supply constraint is not primarily capital: the global reinsurance market has substantial capacity for new lines given current pricing conditions. The constraint is underwriting data. Carriers cannot price AI agent risk accurately without claims experience, and claims experience is sparse because the market is new and most deployments are recent. The result is a cautious market where available products either use performance-data settlement (Munich Re aiSure) to avoid severity uncertainty, require extensive governance documentation to reduce frequency uncertainty, or apply conservative exclusions to limit exposure to the most uncertain scenarios.

The demand signal is real, and the supply side is professionalising alongside it. EIOPA's survey of generative AI in the European insurance sector, published on 2 February 2026 and covering 347 undertakings across 25 countries, found that nearly two thirds are actively using generative AI and that 49 percent now have a dedicated AI policy in place, up from about 25 percent in 2023. Hallucinations and inaccurate outputs were the most cited risk, ahead of cybersecurity and data protection. Those figures describe insurers governing their own use of AI rather than enterprises buying cover, but they matter to a buyer for a practical reason: an underwriter who has formalised AI governance internally has both the vocabulary and the internal mandate to assess it in a client. Enterprise risk managers are meanwhile asking their brokers about AI-specific coverage, and the broker community is beginning to develop specialised AI placement practices. The gap between demand and available products is the market opportunity that several of the carriers below are positioned to capture.

Carrier-by-carrier status: August 2026

Munich Re aiSure

aiSure is the most developed standalone AI insurance product reaching European buyers. Its structure is what distinguishes it: cover attaches when an AI model fails to meet clearly defined performance thresholds, and claims are settled on measurable performance data rather than on proof of a causal chain from AI failure to third-party harm. That makes it materially easier to claim than ordinary liability cover, but it moves the work forward in time, because the performance specification has to be defined and agreed at placement. It is a first-party performance layer. It does not, on its own, answer the claim a harmed customer brings.

On 26 February 2026 Mosaic Insurance announced a partnership under which it underwrites and markets Munich Re's aiSure through its global cyber underwriting team, with up to EUR/USD/CAD 15 million in initial capacity. The stated audience is AI developers and vendors worldwide, which is a narrower group than every enterprise deploying an agent, and a European buyer should read it that way. Munich Re supplies the technical foundation and the underwriting expertise; Mosaic carries the distribution. Access is through Mosaic's cyber specialists rather than a direct approach.

The documentation burden follows from the structure rather than from any published checklist. A performance-triggered product cannot be placed without an agreed, measurable specification for the covered system and a way to evidence performance against it during the policy period. An AI system with no documented performance metrics has nothing for the trigger to reference, which is the practical reason a measurement discipline has to exist before the placement conversation starts.

Armilla AI

Armilla's standalone AI Liability Policy carries limits of up to USD 25 million per organisation and names AI regulatory violations as covered, including defence costs and insurable fines arising from investigations under regulations such as the EU AI Act. That is directly relevant to European enterprises facing enforcement risk under Regulation (EU) 2024/1689, subject to the standing caveat that administrative fines are uninsurable as a matter of public policy in most member states even where defence costs are not. 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, the product having been launched with the Lloyd's underwriter Chaucer.

Armilla's coverage includes third-party errors and omissions liability for AI-generated outputs, regulatory investigation costs for AI Act enforcement actions, and defence costs. Coverage limits reported in public materials reach into the EUR 20 million range for single-enterprise programmes, though limits for smaller enterprises are typically in the EUR 1-5 million range.

Armilla pairs cover with a technical evaluation of the AI system rather than writing on a declaration alone, which is the practical meaning of the coverholder model here: the assessment is what allows delegated authority to be exercised at all. An enterprise without documented governance should expect either a declination or a quote loaded with AI-specific exclusions. Because Lloyd's writes across Europe on a cross-border basis, this London-structured capacity is reachable by EU operators through a broker with the right relationships, even though Armilla is not an EU-domiciled carrier.

Counterpart

Counterpart announced an expansion of its affirmative AI coverage in November 2025, together with a technology errors and omissions insuring agreement, across its miscellaneous professional liability and allied health products. The framing is professional liability rather than management liability: the scenarios addressed are the everyday ones, a consultant delivering a report containing an AI hallucination, or an adviser whose AI-assisted assessment misclassifies an exposure and a client suffers a loss.

Counterpart's endorsements are not standalone AI agent insurance. They sit on the professional liability layer rather than answering for third-party product liability or first-party performance. For an enterprise with a significant agent deployment, Counterpart is a complement to, not a substitute for, dedicated AI cover. It is also a United States product, which for a European reader makes it a wording reference rather than a placement option.

Lloyd's syndicates (bespoke)

Several Lloyd's syndicates are writing bespoke AI liability coverage for larger enterprise programmes. These are structured individually and are not published in product format, which is why no reliable premium floor can be quoted for them. They draw on Lloyd's market capacity and on that market's long habit of writing novel risk lines before a standard product exists. AIUC-1, the standard published by the Artificial Intelligence Underwriting Company, is one of the reference frameworks a governance assessment may be built against.

ElevenLabs secured a first-of-its-kind AI agent insurance policy backed by AIUC-1 certification, which established the first public reference point for how these programmes are structured. The detail of that single placement matters less than the template: an agent is independently assessed against a published standard, and the assessment is what makes an insurer willing to write. Access to bespoke Lloyd's AI programmes requires a specialist broker with Lloyd's market relationships and experience in AI risk placement.

HSB

HSB, part of the Munich Re group, announced AI liability insurance for small and mid-sized businesses on 18 March 2026. The cover is aimed at AI-related losses that may fall outside some general liability policies, including bodily injury, property damage and advertising injury arising from AI-generated advertising, marketing material, blogs and social media content. Two structural facts matter more to a European reader than the coverage grant: HSB does not sell it to businesses directly, it is added to the business policies of carriers that partner with HSB, and availability is subject to insurance regulatory approval.

Its significance here is what it signals rather than what it currently sells in Europe. A Munich Re group entity has concluded that AI liability can be packaged as a standardised small-business product, which is the shape the European market is most likely to grow toward. It is not an EU purchase today. For SME operators comparing this to endorsement-based coverage, see the parallel discussion of accessible AI coverage options in how to choose an AI insurance product on insureyouragent.com.

Testudo

Testudo launched its generative AI liability product on 21 January 2026 as a Lloyd's coverholder, writing for United States enterprises with Apollo capacity behind it. On 26 February 2026 Atrium and QBE joined the panel, lifting the available limit to USD 9.25 million per insured. The policy responds to third-party claims arising from AI-generated outputs, including legal costs and damages, where conventional commercial general liability may not apply.

The practical effect of Testudo and HSB together is to show that capacity providers now see a viable underwriting model below the bespoke enterprise tier. Neither is a European placement today, and both are worth watching for exactly that reason: a product that works at mid-market scale in one jurisdiction is the one most likely to be exported. Any operator evaluating these entrants should confirm coverage scope, exclusions and territory directly, since a lighter underwriting process usually corresponds to a narrower coverage grant rather than an equivalent product at a lower price.

SCOR and Swiss Re: reinsurance market development

SCOR and Swiss Re are not writing primary AI agent insurance for European enterprises, but both are present in the reinsurance layer, and Swiss Re appears among the insurers Armilla names as recognising its programme. Swiss Re Institute has published research on AI risk. The reinsurance question is the one that decides whether this market scales: primary carriers will not launch standardised AI products at volume until they can lay off the tail, and reinsurer engagement is the leading indicator that they will be able to.

The coverage gap: what existing policies typically exclude

The most important market intelligence for most enterprises is not what AI-specific products are available, but what their existing policies already exclude. Three existing lines are most commonly affected.

Errors and omissions (professional indemnity). Many E&O policies written before 2023 contain AI exclusion endorsements added by carriers responding to the rapid expansion of AI deployments. These endorsements typically exclude coverage for any loss arising from the use of AI-generated content or AI-assisted professional services. Where such endorsements exist, they need to be removed or replaced with affirmative AI coverage language before an AI agent incident will be covered.

Cyber liability. Standard cyber policies cover data breaches, ransomware, and network security failures. They do not typically cover AI-specific performance failures that cause financial harm to third parties without involving a data breach. An AI agent that gives incorrect financial advice, approves an ineligible transaction, or discriminates against a protected class in an automated decision produces a loss scenario that cyber coverage is not designed to reach.

Product liability. Product liability policies are being restructured following Directive 2024/2853 (the revised Product Liability Directive, applicable from December 2026), which reclassifies software including AI as a product for strict liability purposes. Policies written under the previous product liability regime may not cover AI software as a product. Post-December 2026, product liability will be a more relevant coverage line for AI deployments, but the transition period and the policy review it requires are a practical compliance task for enterprise risk managers.

For the full analysis of exclusions in cyber and E&O policies as applied to AI agents, see the dedicated article on AI exclusions in cyber and E&O policies on this site. For the regulatory liability framework underlying the coverage demand, see the EU AI Act Article 26 deployer obligations guide.

What did the AI Omnibus change for AI insurance buyers?

Very little, and the fact that it changed very little is the finding. The AI Omnibus entered into force on 27 July 2026 and split the AI Act's schedule in two. The high-risk half moved: Annex III standalone systems to 2 December 2027, Annex I systems embedded in regulated products to 2 August 2028. The other half arrived on time: from 2 August 2026 the transparency obligations including Article 50 and the governance rules apply, and the AI Office and the authorities of the Member States hold responsibility for implementing, supervising and enforcing the Act. The Article 99 penalty regime, up to EUR 15 million or 3 percent of worldwide annual turnover for a qualifying deployer or provider violation on an Annex III system, moves with the obligation it enforces.

If the AI Act deadline had been driving this market, a sixteen-month deferral would have shown up immediately: softer terms, thinner submission requirements, capacity content to wait until 2027 before asking hard questions. None of that happened. Specialist capacity kept expanding, and the submission questions did not change. That is the clearest evidence available that AI liability pricing in Europe is driven by loss pathways rather than by the compliance calendar, and the loss pathways, contractual failure, data protection breach, professional negligence and from 9 December 2026 strict product liability, did not move by a day.

Existing AI-relevant policies continue on their agreed terms until renewal. The practical question for a buyer is what happens at the next one, and for most European buyers the next one falls well before December 2027. The evidence file that becomes a legal requirement then is being asked for at every renewal in the interval. Building it now means building it against a renewal date rather than a statutory one, at your own pace and before every other European deployer is buying the same help in the same quarter.

The deferral also reordered the two deadlines, and almost nobody has re-sequenced their planning for it. Directive 2024/2853, the revised Product Liability Directive, applies from 9 December 2026 and was never in scope of the Omnibus. It is now the nearer deadline by close to a year. Product liability policies are still being restructured in advance of it, and that restructuring, not the AI Act, is the live compliance task for a European enterprise risk manager for the rest of 2026.

Carriers are tightening governance requirements through H2 2026, consistent with what this tracker anticipated in its June and July editions and unaffected by the deferral. Policies binding now are increasingly likely to include warranty conditions requiring that insured high-risk AI systems meet applicable regulatory obligations. A warranty is a stricter mechanism than a disclosure question: a breach can void cover for the connected loss rather than merely adjusting the premium retrospectively. Read new and renewing AI coverage for warranty language specifically. For what applied on 2 August 2026 and what did not, see what actually applied on 2 August 2026; for the underwriting consequence, the Omnibus in force and AI agent insurance; and for the submission itself, what AI insurance underwriters ask before writing a policy.

For enterprises seeking to build the certification evidence that supports both compliance and insurance eligibility, the Agent Certified assessment intake at agentcertified.eu is the starting point. Certification evidence produced before H2 2026 will be most useful for the policy renewals and new placements that enterprises will be pursuing in that period.

Premium benchmarks

Public premium data for AI agent insurance is sparse because most policies are placed through negotiated bespoke programmes rather than standardised product lines. The following benchmarks are derived from disclosed programme details and broker market intelligence as of Q2 2026. They should be treated as indicative rather than definitive.

The governance documentation premium differential, estimated at 30-70% between well-documented and undocumented enterprises for equivalent coverage scope, is the most important financial signal in the current market. It means that investment in AI governance documentation does not just reduce regulatory risk. It has a direct and measurable return in insurance cost reduction at the first renewal following documentation improvement.