How to choose an AI insurance product. A plain-English comparison of the five main options available to operators in 2026.

Why the choice matters now

Until 2025, the AI insurance question for most SMEs was theoretical. Operators assumed their professional indemnity or technology errors and omissions coverage would respond to an AI-related claim in roughly the same way it responded to any professional service claim. Most never tested this assumption.

The cases that have tested it have not produced clean results. The British Columbia Civil Resolution Tribunal's 2024 decision in Moffatt v. Air Canada confirmed that operators are liable for the outputs of their AI agents. The US District Court for the Southern District of New York's 2023 sanctions in Mata v. Avianca established that professional responsibility extends to AI-generated content submitted in legal proceedings. Neither case produced clear insurance coverage outcomes: both produced settlements or outcomes where coverage was either silent or disputed.

The regulatory picture adds urgency to the choice. The EU AI Act's Article 26 imposes explicit deployer obligations on organisations using high-risk AI systems in the EU. The EU Product Liability Directive 2024/2853, applicable from December 2026, extends strict liability to software and AI outputs. Operators in the EU who have not made a deliberate coverage decision before December 2026 are not in a defensible position. The choice of product is a risk management decision, not just a procurement one.

Three factors that determine which product fits

Before comparing individual products, every operator should answer three questions. The answers narrow the field significantly.

1. What type of risk are you carrying?

There are three meaningfully different AI risk types. Performance risk is the risk that the AI produces outputs below a quality threshold that you have committed to contractually: a customer service agent that resolves fewer than the contracted percentage of queries correctly, or a content system whose accuracy falls below a service level. Liability risk is the risk that the AI produces an output that causes a third party to suffer a loss and bring a claim against you. Regulatory risk is the risk of enforcement, fines, or supervisory action arising from non-compliant AI deployment.

Most insurance products cover one or two of these, not all three. Parametric products cover performance risk only. Liability products cover third-party claims but not necessarily regulatory fines, which are often excluded as uninsurable penalties. Understanding which risk type is dominant for your operation is the first narrowing step.

2. What is your AI deployment's autonomy level?

A document summarisation tool that a human reviews before acting is a very different risk from an autonomous agent that takes actions in external systems, communicates with customers, and makes consequential decisions without a human checkpoint. Most underwriters differentiate between these two profiles, even if their policy wordings do not always make the distinction explicit.

Operators whose AI agents act autonomously face a harder underwriting conversation and need products designed specifically for that risk profile. Products designed for AI-assisted decision-making may exclude or limit coverage for fully autonomous action. Know where your deployment sits on the autonomy spectrum before approaching underwriters.

3. What documentation can you produce?

The governance-linked products require documentation that many operators have not yet produced. If you cannot describe your AI system, name the person responsible for oversight, produce a change log, and show an incident management process, the most capable products are not available to you at application stage. This is not a permanent barrier: governance documentation can be built relatively quickly, as the documentation guide on this site explains. But it is a sequencing constraint. Operators buying coverage before building documentation are buying from the products that ask fewer questions, which are also the products with the least favourable terms.

Product 1: Munich Re aiSure

Munich Re's aiSure product, launched in 2024, is the most established parametric AI insurance offering from a major reinsurer. It covers AI performance risk by paying a defined amount when the performance of an AI system falls below a contractually specified threshold. The trigger is metric-based rather than claims-based: if output accuracy, response quality, or another measurable performance indicator drops below the agreed level, the policy pays regardless of whether a third party has made a claim.

The product is most appropriate for operators who have quantifiable performance obligations tied to AI outputs. An operator who has contracted with clients to deliver AI-assisted analysis at a defined accuracy level, or who has built service level agreements around AI-generated outputs, has a natural fit with parametric coverage. The claim settlement is fast and straightforward because there is no need to establish causation or third-party loss.

The limitations are significant. aiSure does not cover third-party liability. If an AI agent produces an output that causes a customer financial loss and the customer brings a claim, aiSure does not respond to that claim. It covers the operator's contractual performance shortfall, not the downstream consequence to the customer. Operators treating aiSure as their primary AI liability coverage are misunderstanding the product's scope. The product is a supplement to liability coverage, not a substitute for it.

Pricing for aiSure is typically structured around the performance threshold, the AI system's measured baseline performance, and the premium paid by the operator for exceeding that threshold. Munich Re calculates premiums based on proprietary performance benchmarking. Operators should expect a meaningful underwriting process involving the AI system's actual performance data, not a commodity premium calculation.

Product 2: AIUC-1 backed policies

Artificial Intelligence Underwriting Company (AIUC) was founded in 2024 with backing including Nat Friedman's NFDG fund, Emergence, and Terrain. In 2025, AIUC published the AIUC-1 standard, the first underwriting framework specifically designed for agentic AI systems. ElevenLabs became the first policyholder under the AIUC-1 standard in February 2026.

The AIUC-1 framework is built around adversarial evaluation. Before an AI system is underwritten under the standard, it is subjected to a programme of simulated attack scenarios: hallucination probing, data leakage attempts, IP boundary testing, harmful content generation attempts, and faulty tool action scenarios. The adversarial simulation programme is reported to include more than 5,000 individual test scenarios per system evaluation. Coverage is only available to systems that pass this evaluation at a defined threshold.

The coverage categories under AIUC-1 include hallucination liability (claims arising from factually incorrect AI outputs), data leakage (claims from unintended disclosure of confidential or personal data by the AI), intellectual property infringement (claims arising from AI outputs that reproduce protected content), harmful content (liability from AI-generated outputs that cause psychological or reputational harm), and faulty tool actions (liability from incorrect actions taken by autonomous AI agents in external systems).

The product is the most technically appropriate for operators deploying autonomous or semi-autonomous agents that take real-world actions. The evaluation process is rigorous and time-consuming, which means AIUC-1 backed coverage is not a product an operator can obtain in the week before a deployment. Build six to twelve weeks of lead time into any project plan that requires AIUC-1 coverage.

The primary limitation from a European perspective is regulatory alignment. The AIUC-1 standard was designed primarily for the US market and does not produce the technical documentation required under Annex IV of the EU AI Act, which applies to high-risk AI systems. European operators whose deployments fall within the EU AI Act's high-risk categories need to treat AIUC-1 as one layer of coverage and maintain separate EU AI Act compliance documentation alongside it. The full analysis of AIUC-1 and its European implications on agentinsured.eu covers this gap in detail.

Product 3: Armilla

Armilla operates as a Lloyd's of London coverholder, meaning it underwrites policies backed by Lloyd's capacity. It offers technology errors and omissions coverage for AI systems, with per-incident limits up to $25 million. Armilla's differentiating feature is that coverage is directly linked to the operator's governance documentation: the quality and comprehensiveness of the operator's AI governance determines both eligibility and premium.

The governance documentation Armilla requires at submission includes an AI use policy, an incident response plan, a documented risk management process, and evidence that human oversight is in place for consequential AI decisions. Operators who can produce all of these at a meaningful level of detail typically receive more favourable terms than operators who cannot. Operators with no governance documentation are unlikely to obtain coverage at standard market rates.

The coverage scope under Armilla's product covers third-party losses arising from AI errors and omissions, including incorrect outputs that cause financial loss, failure to perform as specified, and outputs that breach professional standards where the operator is a regulated professional service. The product is most appropriate for professional services firms, technology companies, and operators whose AI systems support decisions with significant downstream financial or legal consequences.

The governance linkage means Armilla is simultaneously an insurance product and an incentive to build proper AI governance. Operators who build governance to obtain Armilla coverage are building documentation that also satisfies EU AI Act deployer obligations under Articles 26 and 17. This overlap is not accidental: the insurer's interest in governance-documented operators and the regulator's interest in governed deployments are aligned, and operators who understand this can build one programme that serves both audiences.

Product 4: Counterpart Affirmative AI Coverage

Counterpart, a US specialty insurer focused on management liability and technology E&O, introduced its Affirmative AI Coverage in November 2025. The product differs from standalone AI insurance in that it adds explicit, affirmative AI coverage language to existing management liability and technology errors and omissions policies, rather than offering a separate AI policy.

The affirmative structure is significant. Most technology E&O policies were written when AI liability was not a forethought of the underwriting process. Coverage for AI-related losses sits in a grey zone between what the policy explicitly covers and what it explicitly excludes. Counterpart's approach is to remove the ambiguity: coverage for AI-related losses is either explicitly included or explicitly excluded, with no silent grey zone. For operators who have existing technology E&O or management liability with Counterpart, the add-on provides clarity about what is and is not covered for AI-related losses without requiring a separate placement.

The product is most appropriate for operators who already have Counterpart technology E&O or management liability in place, and who want to resolve the AI coverage ambiguity without undertaking a full standalone AI placement. For operators without existing Counterpart coverage, the incremental value of affirmative AI language is more limited since the underlying policy placement is still required.

The coverage scope focuses on third-party claims arising from AI outputs that cause financial or reputational harm, including claims arising from AI-generated professional advice, AI-assisted decisions in employment or lending, and AI-produced communications that make misrepresentations. The product does not cover regulatory fines or penalties, which are excluded as uninsurable in most jurisdictions.

Product 5: Traditional coverage extended by endorsement

The most widely available option for SMEs with limited underwriting relationships is to extend an existing professional indemnity, technology E&O, or general liability policy by endorsement to include AI-related losses. This is the least technically rigorous option and often the most budget-accessible.

The limitations are substantial. Traditional policies were not designed for AI risk, and the endorsement coverage available from most standard market insurers reflects this: exclusions for intentional AI use, for AI systems operating without human oversight, and for AI systems that have been materially modified since the policy was placed are common. The policy wording may create ambiguity about whether a specific AI-related loss is covered, which puts the operator in a worse position than explicit coverage or explicit exclusion.

Operators considering the endorsement route should ask their broker for the full endorsement wording, not just a summary. They should confirm in writing which specific AI use cases are covered, what the coverage limit is for AI-related losses (which may be sub-limited below the main policy limit), and what documentation the insurer requires to maintain coverage in force. Without these answers in writing, the endorsement provides the appearance of coverage without the substance.

Traditional endorsement coverage is most appropriate as a transitional option for operators who are building governance documentation with a view to obtaining a more suitable product at the next renewal cycle. It is not recommended as a long-term coverage solution for operators whose AI agents are central to their service delivery or whose deployments involve significant autonomy.

Decision matrix: matching product to operator

The following framework summarises how to match the five product categories to the most common SME operator profiles.

Operators with contractual performance obligations for AI output quality and no significant third-party liability exposure: Munich Re aiSure parametric coverage is the most direct match. Add third-party liability coverage alongside it if customer interactions are involved.

Operators deploying autonomous or semi-autonomous AI agents that take actions in external systems, communicate with customers, or make consequential decisions: AIUC-1 backed coverage is the most technically appropriate. Allow six to twelve weeks for the adversarial evaluation process. European operators add EU AI Act documentation alongside.

Professional services firms, technology companies, and operators whose AI supports regulated decisions or high-value advice, with at least basic governance documentation in place: Armilla is the most appropriate match. Build governance documentation to the level required before approaching underwriters.

Operators who already have Counterpart management liability or technology E&O and who want to resolve AI coverage ambiguity without a standalone placement: Counterpart Affirmative AI Coverage is the most efficient option.

Operators at early stage, without governance documentation, with low AI autonomy, and on a constrained budget: Traditional endorsement as a transitional option, with a plan to move to a purpose-built product within twelve months. Use the transition period to build governance documentation.

Questions to ask your broker before placing

Before committing to any AI insurance product, four questions should be answered in writing by your broker. First, which specific AI use cases are explicitly covered under the policy, and are any use cases explicitly excluded? Second, what is the coverage limit for AI-related losses, and does it differ from the main policy limit? Third, what documentation does the insurer require to maintain coverage in force, and what happens if a material change to the AI system is not notified? Fourth, does the policy cover third-party claims arising from AI outputs, or only performance shortfalls on contractual obligations?

If your broker cannot answer these questions specifically, consider engaging a broker who specialises in technology and AI risk. The AI insurance market is developing faster than the general brokerage market's familiarity with it, and a broker without specific expertise in AI coverage is likely to place you in a product that fits the general category but misses the specific risks you are running.

For a step-by-step pathway from coverage decision to actual placement, the coverage pathway on this site covers the three stages: documentation, submission, and policy selection. For a broader view of which documentation the most capable products require, the five questions before deploying an AI agent guide covers the pre-coverage diagnostic in full.