Counterpart affirmative AI coverage. What it covers, what it excludes, and how European enterprises access it.

In November 2025, Counterpart Insurance became one of the first US insurers to launch explicitly affirmative AI coverage as a standalone endorsement. Rather than relying on the absence of an exclusion, the product names AI failures as covered perils in their own right. This analysis reads the coverage structure, examines what enterprises need to do to qualify, and maps the access options for European companies that cannot reach Counterpart directly.

What Counterpart's affirmative AI coverage is

Counterpart is a technology-focused insurtech that distributes management liability and professional liability products in the US market. Its primary products are Directors and Officers (D&O) and Employment Practices Liability (EPL), alongside Miscellaneous Professional Liability (MPL) and Technology Errors and Omissions (Tech E&O). In November 2025, it added an Affirmative AI Coverage endorsement to its MPL and Allied Health products, with a Tech E&O insuring agreement.

The product was developed in response to a market problem that had become commercially uncomfortable. Standard professional liability and tech E&O policies were written before AI agents became consequential participants in professional advice, healthcare triage, and hiring decisions. Their language either excluded AI-generated losses by name, or contained ambiguous technology-related exclusions that left AI failure claims in an unsettled coverage position. When insureds suffered losses from AI failures and sought to claim, the coverage dialogue was slow, adversarial, and frequently resulted in dispute.

An affirmative endorsement solves this by naming the risk as a covered peril. The insured does not need to argue that an AI hallucination causing professional loss is covered. The policy says it is. This shifts the coverage question from "is this excluded?" to "is this within the agreed trigger conditions?" which is a much cleaner claims process.

What the three trigger categories cover

Counterpart's endorsement as launched names three categories of AI failure as covered triggers. Each category covers loss caused by the trigger, subject to the policy limits and deductibles.

AI hallucinations. A hallucination is an output from a generative AI system that is factually incorrect, invented, or presented with false confidence. In professional services contexts, a hallucination that a client acts on can generate significant loss: incorrect legal citations (as in Mata v. Avianca, SDNY 2023), fabricated medical references, invented financial product terms, or incorrect regulatory guidance. Counterpart's hallucination trigger covers financial loss caused by a client or third party acting on a hallucinated output from an AI system the insured deployed.

AI misclassification. Misclassification covers errors in AI-driven categorisation decisions: credit risk scores, fraud classifications, insurance underwriting outputs, or any context where the system assigns a person, transaction, or object to the wrong category with consequential effect. The practical application in financial services is a credit decision that was refused or mispriced because the AI model misclassified the applicant's risk profile. Loss flowing from that misclassification, including customer complaints, regulatory fines, or reputational damage, is within the covered category.

AI-driven hiring bias. This trigger addresses the Employment Practices Liability exposure created by AI hiring tools that produce discriminatory outcomes. It covers claims from candidates who were rejected or disadvantaged by an AI screening or ranking system that incorporated discriminatory proxies, whether or not the employer intended the bias. The coverage connects to the growing enforcement environment around hiring AI: EEOC guidance in the US and emerging EU AI Act obligations for employers using AI in recruitment both create pathways for regulators and claimants to pursue these losses.

What it does not cover

Understanding the exclusions is as important as understanding the covered triggers. Counterpart's affirmative AI endorsement does not cover losses arising from AI systems the insured knew, or reasonably should have known, were operating outside their documented parameters. This creates a significant incentive for robust pre-deployment testing and monitoring: an insured who deployed an AI system without bias testing and is then found to have known the system produced discriminatory outcomes is unlikely to receive coverage for the resulting claims.

The endorsement does not cover losses from AI systems deployed in medical diagnostics or mental health applications, consistent with Armilla's exclusion in the same areas. The systemic risk and governance complexity of clinical AI remains outside the standard affirmative coverage category. Bespoke coverage for clinical AI exists through specialist health tech markets but is not within Counterpart's current product scope.

Physical injury losses, property damage, and losses from AI systems used in critical infrastructure contexts are not within the MPL and Tech E&O framework. These risks may fall under other product lines or remain uninsured. For an overview of the standard exclusion architecture across all AI-adjacent product lines, see the AI exclusions guide for cyber and E&O policies.

The governance requirements for qualification

Counterpart's underwriting process requires applicants to complete an AI-specific supplement to the standard MPL or Tech E&O application. The supplement is structured around governance evidence. Enterprises that cannot complete it with real documentation rather than aspirational statements are unlikely to receive the endorsement, or will receive it at premium levels that reflect the underwriter's view of undisclosed governance risk.

The practical documentation required maps to five categories. An AI system inventory that names each AI system deployed, its intended use, the provider or foundation model it relies on, and the population of decisions it affects. A risk management record for each inventoried system that describes the pre-deployment assessment, the residual risk identified, and the controls applied. Evidence of bias testing, either through the provider's documentation or through independent testing the applicant conducted. A named executive accountable for AI governance, with a brief description of the governance structure they oversee. An incident response procedure that covers how AI failures are identified, escalated, and documented.

Enterprises that have prepared for EU AI Act compliance under Articles 9, 26, and 27 of Regulation (EU) 2024/1689 will find significant overlap between what the regulation requires and what Counterpart's underwriting supplement asks for. The EU compliance file is not identical to the US underwriting submission, but it covers most of the same ground. For the connection between EU AI Act compliance documentation and insurance eligibility, see the overview on Article 26 deployer obligations.

The claims trigger and the AIUC-1 comparison

The claims trigger architecture of Counterpart's endorsement is occurrence-based within the policy period: a loss must be caused by a covered AI failure that occurred during the policy period. This is distinct from the parametric structure used by Munich Re aiSure, which settles claims on measured performance data without requiring causal proof of a specific incident. Both approaches resolve the same underlying problem but through different mechanisms.

AIUC-1, the certification standard published by the Artificial Intelligence Underwriting Company in 2025 and used to structure ElevenLabs' February 2026 coverage, takes a third approach. AIUC-1-backed policies tie coverage terms and pricing to the certified behaviour of the AI agent: a system that passes AIUC-1 assessment receives coverage structured around its documented behaviour, and the underwriting relies on the certification rather than on a post-incident causal analysis. This produces a cleaner coverage dialogue but requires the insured to go through the AIUC-1 certification process before coverage can be offered.

Counterpart's endorsement does not require AIUC-1 certification, which makes it more accessible to enterprises that have not yet engaged with the certification process. The tradeoff is a more complex claims process, because each claim requires causal proof that the loss was caused by one of the named AI failure triggers rather than by another factor.

How European enterprises can access comparable coverage

Counterpart is a US company. Its standard distribution is through US-licensed brokers and managing general agents. European enterprises domiciled outside the US cannot access Counterpart's endorsement through standard direct placement. Lloyd's surplus lines placement is theoretically available but involves an additional intermediary layer and is not a direct-access route for most European compliance teams.

For European enterprises seeking affirmative AI coverage for the same trigger categories, two providers offer the closest functional equivalent. Armilla, as a Lloyd's coverholder backed by Chaucer and Axis Capital, can write coverage for European enterprises through Lloyd's broker channels. Coverage limits reach USD 25 million per company following Armilla's January 2026 capacity expansion. Armilla's evaluation process requires technical testing of the AI system, governance documentation review, and excludes medical diagnostics and mental health applications, consistent with Counterpart's exclusions.

Munich Re aiSure offers parametric AI performance insurance that covers measurable performance failures, including accuracy degradation, output quality failure, and defined performance SLA breaches. Unlike Counterpart's occurrence-based trigger, aiSure settles on measured data, which removes the causal proof requirement from the claims process. Coverage is available to European enterprises through Munich Re's European market presence.

For a full map of the AI insurance providers available to European enterprises and their coverage structures, see the AI liability insurance market map for 2026. For the certification pathway that positions an enterprise for preferential rates with any of these providers, see the FP Certified methodology on the certification platform.

What the Counterpart launch signals about the market

The November 2025 launch of Counterpart's affirmative AI coverage, alongside AIUC-1's first backed policy with ElevenLabs in February 2026, marks a structural inflection in the AI insurance market. Two years ago, affirmative AI coverage did not exist as a named product category. By the end of 2025, at least four providers were actively writing it. The market that AIUC estimated at near-zero in 2023 is growing toward the $500 billion category projection they published for 2030.

For European enterprises, the practical implication is that the window in which "no AI-specific coverage exists" is closing. By mid-2027, when EU AI Act enforcement is fully normalised and the Product Liability Directive brings AI software within strict product liability, European insurers will almost certainly have launched domestic affirmative AI coverage products. The enterprises that have spent 2026 building governance documentation and engaging with the existing market will be positioned for preferential terms. Those who have not will be negotiating coverage from a weaker position under more urgent compliance pressure.