Can you insure an AI agent that has not been certified yet?

Why this question gets asked more since enforcement began

Since 2 August 2026, when the EU AI Act's Article 50 transparency obligations began to apply, and with Annex III high-risk obligations now due from 2 December 2027 under the AI Omnibus, Regulation (EU) 2026/1744, in force since 27 July 2026, enterprises that had been deferring a decision on AI agent coverage are now approaching brokers under time pressure, and a large share of them have not completed any formal governance assessment. The natural assumption, reinforced by how AIUC-1 and Agent Certified are discussed as underwriting evidence throughout this network, is that certification is a locked door: no certificate, no quote. That is not accurate for the market as a whole, though it is close to accurate for a specific and important segment of it, which this article separates out carrier by carrier.

Where certification is genuinely close to a gating requirement

Armilla AI, a Lloyd's coverholder whose affirmative AI liability insurance is underwritten by certain underwriters at Lloyd's, discussed in full in the Armilla coverage guide on this site, operates the most demanding underwriting posture in the market, with the AIUC-1 standard forming the baseline for its assessment process. Armilla's coverage extension to EU AI Act regulatory violations, the product feature most relevant to an operator facing enforcement risk, is specifically the part of the offering most dependent on governance evidence, because regulatory liability coverage requires the underwriter to form a view on how likely a compliance failure actually is. Enterprises approaching Armilla without documented governance, certification-grade or not, typically face declinature or coverage issued with significant AI-specific exclusions rather than a straightforward decline. The same applies to the Lloyd's syndicates writing bespoke enterprise programmes under AIUC-1 reference, where minimum premium thresholds above EUR 100,000 already imply an enterprise-scale submission that these underwriters expect to be accompanied by enterprise-scale evidence.

It is worth being precise about what "close to gating" means here. It is not that Armilla or an AIUC-1-referenced syndicate has a written policy refusing all uncertified submissions. It is that their underwriting process is structured around exactly the evidence a certification assessment produces, so an operator without a certificate is, in practice, being asked to reconstruct the equivalent evidence from scratch during underwriting, under time pressure, with an underwriter rather than an independent assessor forming the first judgement on its adequacy. Operators who have gone through this process report it is materially slower and less predictable than arriving with a completed assessment.

Where certification helps but is not required

Munich Re's aiSure product settles against a defined performance specification, which places less emphasis on governance documentation and more on a defined performance specification against which coverage attaches. An operator can secure aiSure coverage without a formal certification by providing a clear specification of the AI system's intended performance and a description of pre-deployment testing, which is a lighter evidentiary bar than Armilla's governance-first approach, though certification evidence still strengthens the specification quality and can support better terms.

Counterpart's affirmative AI endorsements, on its miscellaneous professional liability and allied health products with a technology E&O insuring agreement, are explicitly designed as the most accessible entry point in the market for enterprises without a formal AI governance programme. These endorsements do not require certification and were built for exactly the operator segment this article addresses: a business that knows it has AI liability exposure but has not yet built a governance evidence file.

HSB, a Munich Re subsidiary, introduced AI liability insurance for small and mid-sized businesses in March 2026, added to partner carriers' business policies rather than sold direct, with a simplified underwriting process focused on a description of the AI system's function, the population it interacts with, and confirmation of basic oversight measures such as a named responsible person. Testudo, a Lloyd's coverholder writing for United States enterprises since 21 January 2026, with Apollo capacity joined by Atrium and QBE, occupies similar territory. Neither requires certification.

The real evidence underneath every questionnaire

What separates an operator who gets a workable quote without certification from one who does not is rarely the presence or absence of a certificate itself. It is whether the operator can produce, in some form, the same underlying evidence a certification assessment reviews: what the AI system does and its known limitations, who is responsible for oversight and how that oversight actually functions, what logging exists, what the provider's instructions for use say, and what the incident response process looks like. This is the same Article 26 operator file described across this network's coverage of EU AI Act compliance, and it is worth stating plainly here: a carrier's underwriting questionnaire is, in substance, asking for this file whether or not the word certification ever appears in the submission form.

An operator who has built this file to a genuinely defensible standard, even without an independent assessor's score attached to it, is in a materially stronger underwriting position than one with no documentation, certified or not. This is the practical reason certification and insurability are so closely linked in this market without certification being formally mandatory: certification is simply the fastest and most portable way to demonstrate the same evidence a good self-assembled file also demonstrates, verified by an independent party rather than asserted by the operator alone.

The pricing consequence of going in uncertified

Where coverage is available without certification, it is very rarely available at the same price as an equivalent certified submission. The governance documentation premium differential observed across the European market, drawn from disclosed programme details and broker market intelligence, is estimated at 30 to 70 percent between well-documented and undocumented enterprises for equivalent coverage scope. A completed certification is the single artefact that satisfies the documentation requirement across nearly every carrier simultaneously, which is why the relative benefit is largest for an operator seeking multiple competing quotes rather than a single placement with one accessible carrier.

The practical implication is not that an uncertified operator should delay buying any cover until certification is complete. It is that the cost of delay should be weighed against the cost of the pricing differential, and for most operators facing an immediate compliance or exposure deadline, binding accessible cover now and certifying before the next renewal is the more defensible sequence than leaving an active AI deployment uninsured while an assessment is pending.

How to sequence a first placement

The right order depends on how much time is available. An operator with several months before a renewal, a compliance deadline, or a specific enterprise counterparty request for governance evidence should pursue certification first, because it widens market access to include Armilla and AIUC-1-referenced Lloyd's syndicates and produces materially better terms across the accessible options as well. The Agent Certified assessment intake is the starting point for that path.

An operator that needs cover immediately, because a deployment is already live and uninsured, should approach the accessible options first: HSB, through its partner carriers, for small and mid-sized businesses, Testudo for United States enterprises, or Counterpart for an endorsement on its miscellaneous professional liability and allied health cover, all covered in the full European AI agent insurance market tracker on this site. None of these require certification to bind. Certification can then follow to improve terms at the next renewal, using the same evidence gathering exercise that the initial submission already required in a lighter form.