AXA, Allianz, Zurich: where the tier-one European insurers stand on AI liability in 2026

The three largest European primary insurers, AXA, Allianz and Zurich, have not launched dedicated AI agent liability products as of mid-2026. Their public positions acknowledge AI as a material emerging risk. Their commercial lines offerings do not yet translate that acknowledgment into coverage. This analysis maps what each carrier has said publicly, why they have held back, and what the resulting gap means for European enterprises now operating AI systems under Regulation (EU) 2024/1689 and Directive 2024/2853.

Why the tier-one position matters

In any European commercial insurance market, the tier-one primary carriers set the baseline. AXA, Allianz and Zurich collectively write the majority of commercial liability premiums for large and mid-size European enterprises. When they include a risk class in their standard products, that class becomes insurable for most of their client base without specialist broker involvement. When they exclude it or remain silent on it, the enterprise risk management function faces a structural coverage gap that requires active remediation through specialist channels.

For AI agent liability, the tier-one carriers have been in an observation posture since 2023. Their research and analytics functions have produced substantive published work. Their specialty units have handled individual large-risk placements. But their standard commercial lines products have not moved. Understanding why, and understanding what that means for the enterprise coverage gap, requires examining each carrier's position individually.

AXA: research-forward, product-cautious

AXA has one of the most developed public analytical positions on AI risk among European primary carriers. The AXA Research Fund has supported academic work on AI safety and liability since the mid-2010s, and AXA Group Risk Management has identified AI liability in its annual emerging risk reports as a top concern. AXA XL, the specialty and large-risk reinsurance division, has been the most commercially active part of the group in discussions with large enterprise clients about AI-related exposures.

The AXA position on AI in commercial lines has been to integrate AI risk questions into existing product frameworks rather than to launch standalone products. AXA's cyber and technology insurance underwriting teams in France, Germany and the United Kingdom have developed AI-specific questionnaire sections for submission documents. Underwriters ask about AI governance frameworks, model validation processes, human oversight arrangements, and EU AI Act compliance status for clients in scope. However, the policy wordings that result from this more informed underwriting process still sit within AXA's standard cyber and professional indemnity structures, with exclusions that limit coverage for autonomous system errors to scenarios the underwriter specifically endorses.

AXA's public communications on AI liability in 2025 and early 2026 have consistently referenced the EU AI Act as a framework AXA is monitoring, rather than one it is already pricing into products. This posture reflects a deliberate choice: wait for the regulatory framework to stabilise before building products priced against its obligations. The Digital Omnibus provisional agreement of 7 May 2026, which may defer the 2 August 2026 high-risk enforcement deadline to 2 December 2027, has if anything reinforced this posture. The original deadline binds until the Digital Omnibus is formally adopted into EU law, but AXA's commercial lines teams are unlikely to launch products timed to a deadline that may not be enforced on schedule.

The practical consequence for AXA commercial clients is that AI agent liability coverage is currently available only through AXA XL specialty placements for large risks, or through endorsement negotiation with AXA's professional liability unit on a case-by-case basis. No product is available through standard AXA commercial lines channels that affirmatively covers AI agent errors, autonomous decision outputs, or model performance failures.

Allianz: the most analytically active, still no dedicated product

Allianz Global Corporate and Specialty (AGCS) publishes what is arguably the most detailed annual AI risk analysis of any European primary carrier through the Allianz Risk Barometer. AI and digital transformation risks have ranked in the AGCS top-ten risk survey results since 2023, and the 2025 and 2026 barometers have specifically called out AI liability, AI model failure, and AI regulatory breach as sub-categories of commercial risk that enterprise risk managers are prioritising.

AGCS has also published client-facing guidance on AI risk management expectations. Their published AI risk guidance outlined the documentation standards AGCS expects from large commercial clients deploying high-risk AI systems, covering governance frameworks, model testing documentation, incident response procedures, and regulatory compliance evidence. This guidance is consistent with the EU AI Act deployer obligations under Articles 9 to 17 and 26 of Regulation (EU) 2024/1689 and with the risk engineering expectations documented in the Swiss Re and Munich Re AI underwriting frameworks.

Despite this analytical depth, Allianz has not launched a named AI agent liability product through standard commercial lines. The Allianz technology errors and omissions and cyber products include AI-related endorsements that can be negotiated, but no standalone AI agent liability product is available in the Allianz standard product range as of June 2026.

Allianz Re, the reinsurance division, is understood to be developing internal accumulation models for AI liability exposure across the Allianz group. The construction of an accumulation model is a prerequisite for writing AI liability business at scale: without understanding aggregate exposure across a client portfolio, a carrier cannot price or limit the risk prudently. The fact that accumulation modelling is underway suggests Allianz is progressing toward a product, but the timeline for a public launch has not been disclosed.

For AGCS clients and mid-market Allianz commercial clients, the current position is similar to AXA's: AI agent liability coverage requires specialist placement or endorsement negotiation. The AGCS AI risk documentation guidance is useful as a preparation framework because it clarifies what AGCS underwrites expect to see. Enterprises that produce that documentation are better positioned for both the endorsement negotiation and for eventual dedicated product access.

Zurich: integration-led approach without a named AI product

Zurich Insurance Group has approached AI risk primarily through its risk engineering and risk advisory functions rather than through new product launches. Zurich Risk Engineering has published AI risk assessment frameworks and has provided AI risk advisory services to large commercial clients as part of its broader risk engineering offering. Zurich's published risk engineering guidance articulates the governance and control expectations Zurich applies when assessing AI deployments in commercial clients.

Zurich's commercial lines products, including their technology, cyber and professional indemnity offerings, have been updated to include AI-specific questions in submission processes, and Zurich's underwriters have authority to add AI-related endorsements to existing policies for clients where the risk engineering assessment supports it. Zurich Resilience Solutions, the risk advisory and engineering business, has developed AI risk scoring tools for large enterprise clients that produce assessments aligned with the EU AI Act high-risk classification framework.

However, Zurich has not launched a dedicated AI agent liability product. Their approach mirrors Allianz's in prioritising client risk management quality over product launch speed. The Zurich position, as articulated in investor briefings and risk engineering publications, is that AI liability is best addressed through improving client governance standards rather than through broad product availability that may create adverse selection.

The adverse selection concern is one that all three carriers share. In a market where loss history is thin and enterprise AI governance standards vary enormously, a broad product launch without strict underwriting criteria risks attracting the clients with the worst AI governance and the highest loss probability. The specialist MGA model (Armilla, Counterpart) and the reinsurer-backed product model (Munich Re aiSure) address this by building detailed governance documentation requirements into the underwriting process. The large primary carriers, whose standard commercial lines distribution does not naturally select for that level of documentation, are more exposed to adverse selection risk on a broad product launch.

The EIOPA survey and the supervisory view

The February 2026 EIOPA survey on AI in the European insurance sector provided the most authoritative external view of where the market stands. EIOPA surveyed European insurers on their AI use in underwriting and their AI-specific product development. The survey results, published by EIOPA in its February 2026 report on AI in insurance, confirmed that no large European primary insurer had yet launched a dedicated AI liability product comparable to the specialist MGA offerings in the US and Lloyd's market.

EIOPA described the European AI insurance market as "nascent" and identified three barriers consistent with what this analysis finds at AXA, Allianz and Zurich: the absence of a standardised risk assessment framework, the thin loss data environment, and the uncertain regulatory timeline. EIOPA also noted that European supervisors are watching whether the AI Act enforcement process generates claims activity that can anchor actuarial pricing, which further reinforces the large carriers' wait-and-observe posture.

For European enterprises, the supervisory acknowledgment that the market is nascent provides important context for risk management decisions. The gap is not a gap that will close quickly through normal commercial lines channels. It requires deliberate engagement with the specialist market that currently has coverage available, using the same documentation and certification approach that positions enterprises for access to tier-one products when they eventually launch.

The EIOPA survey findings are analysed in detail in the dedicated EIOPA survey analysis on this platform. That article provides the full breakdown of insurer readiness ratings and the EIOPA recommendations for product development timelines.

What the gap covers: regulatory liability exposure without a primary policy response

The coverage gap is not abstract. It maps to specific regulatory liability channels that are now active or becoming active for European enterprises.

Under Regulation (EU) 2024/1689, operators of high-risk AI systems are obligated to implement risk management systems (Article 9), data governance controls (Article 10), technical documentation (Article 11), logging and record-keeping (Article 12), transparency measures toward users (Article 13), human oversight mechanisms (Article 14), and quality management systems (Article 17). Failure to meet these obligations creates supervisory and civil liability exposure. A regulator finding that an enterprise failed to maintain adequate human oversight under Article 14 can support a civil claim by a user harmed by the AI system's output. Standard AXA, Allianz and Zurich policies are not structured to respond to that liability channel, because the coverage trigger in a standard professional indemnity policy is a professional act with a human author, not a system output with a regulatory compliance shortfall.

Under Directive 2024/2853 on liability for defective products, which will apply from December 2026, AI systems embedded in products can create product liability exposure for the manufacturer or importer. The directive extends the product liability concept to software and AI in ways the 1985 Product Liability Directive did not. An AI agent embedded in a software product that causes damage to a user can trigger product liability under Directive 2024/2853 on grounds that the AI component was defective. Standard product liability policies from AXA, Allianz and Zurich are designed around physical product defects, not software or AI performance failures.

The Moffatt v. Air Canada case (British Columbia Civil Resolution Tribunal, 2024) established a relevant precedent: an operator cannot escape liability for its AI agent's outputs by arguing the AI acted independently. Air Canada argued its chatbot was a separate legal entity whose statements did not bind the company. The tribunal rejected that argument. The practical consequence for European enterprises is that AI agent errors in customer-facing interactions create direct operator liability regardless of whether a human reviewed the output. This liability channel sits outside the standard professional indemnity coverage trigger in all three carriers' current wordings.

In the US context, Mata v. Avianca (SDNY, 2023) illustrates the professional liability dimension: an attorney's use of an AI-generated brief containing fabricated citations created professional liability exposure for the law firm. European professional liability standards do not require a US case to establish the principle. Under standard professional indemnity coverage from European carriers, whether the AI-generated output is treated as a covered "professional act" depends on the specific wording and the level of human review, and most current wordings do not affirmatively address AI-generated outputs.

Where coverage is currently available

While AXA, Allianz and Zurich are in an observation posture, coverage for AI agent liability is available through four channels that European enterprises can access now.

Munich Re aiSure is the most structured European-connected AI insurance product available as of mid-2026. Distributed through the Mosaic Insurance partnership, aiSure offers parametric coverage triggered by defined AI performance failures. The parametric structure requires defined coverage triggers, which in turn requires the insured to specify their AI system's performance benchmarks at underwriting. Enterprises that can articulate measurable performance standards, including those required under EU AI Act Article 9 risk management processes, are well positioned for aiSure coverage.

Armilla's Lloyd's-backed programme provides affirmative AI liability coverage with backing from Chaucer and Axis Capital. Armilla's underwriting process focuses on the enterprise's AI governance documentation, model testing evidence, and vendor evaluation practices. The coverage structure is closer to a professional indemnity form than a parametric product, providing third-party liability coverage for AI agent errors that cause financial loss to end users or counterparties. Armilla's January 2026 capacity expansion extended coverage availability to European enterprises operating in regulated sectors.

Counterpart's affirmative AI coverage product is available primarily in the US market but can be accessed by European enterprises with US operations or through London-market specialty brokers. The Counterpart structure uses an affirmative AI endorsement to a professional liability base policy, explicitly including rather than excluding AI-generated outputs within the coverage grant.

HSB, the specialty insurer within Munich Re Group focused on equipment and technology risk, has developed technology reliability products that can cover AI system failures in operational technology environments; availability for European clients varies and should be confirmed with a specialist broker. HSB's coverage is most relevant to industrial AI deployments and AI-embedded operational equipment rather than to enterprise software or customer-facing AI agent deployments.

For detailed coverage structure comparisons across these products, see the AI liability insurance market map on this platform.

What enterprises should do while the tier-one gap persists

The tier-one gap will close. The regulatory enforcement timeline for the EU AI Act, even if the Digital Omnibus deferral is formally adopted, creates a claims environment that will eventually produce the loss history large carriers need to price AI liability products. Directive 2024/2853 will be applicable from December 2026, and its extension of product liability to AI-embedded software will generate cases. The enterprise AI deployment population is scaling rapidly, and with scale comes the adverse selection-resistant volume that makes standard commercial lines products commercially viable. The question is not whether AXA, Allianz and Zurich will launch AI liability products, but when.

Between now and then, enterprises face three practical tasks.

The first is documentation. The governance and technical documentation that EU AI Act Articles 9 to 17 and 26 require is the same documentation that Munich Re aiSure, Armilla, and AGCS specialty underwriters ask for when evaluating an AI liability risk. Producing that documentation for regulatory compliance produces it for insurance purposes simultaneously. Enterprises that have not yet built a systematic EU AI Act documentation programme are accumulating both regulatory exposure and insurance inaccessibility at the same time.

The second is independent certification. An Agent Certified assessment produces an evidence file structured against the governance dimensions that both the EU AI Act and specialist AI liability underwriters evaluate. Independent certification converts a self-reported governance claim into third-party-verified evidence, which materially changes an underwriter's confidence in the risk. When tier-one products launch, the underwriting criteria will almost certainly require governance documentation of this quality.

The third is market access now. Enterprises with current AI liability exposure, particularly in regulated sectors, cannot wait for tier-one product launches that may be 12 to 24 months away. Accessing Munich Re aiSure, Armilla or Counterpart coverage now through a specialist broker with European AI market relationships is the appropriate risk management response to a gap that is structural rather than temporary.

To join the pipeline for structured coverage guidance as the market develops, see the Agent Insured waitlist.