Do I need to disclose my AI agents when applying for insurance?

Why this question matters more than it used to

Five years ago, an enterprise applying for commercial insurance had little reason to think about whether an internal automation tool needed to be mentioned to an underwriter. AI agents that interact directly with customers, make consequential decisions, and take autonomous action have changed that. The question of whether an AI agent must be disclosed at underwriting is no longer academic, because the answer determines whether a policy actually responds when that agent causes a loss.

The starting point is not AI-specific law. It is the ordinary pre-contractual disclosure duty that has underpinned commercial insurance contracts for centuries, applied to a new category of material fact.

The disclosure duty, in plain terms

Commercial insurance contracts are contracts of the utmost good faith. Before a policy is bound, the party seeking insurance owes the insurer a duty to disclose facts that are material to the risk, meaning facts that would influence a prudent underwriter's decision to write the risk or the terms on which it would do so. This duty exists because the insurer is pricing a risk it cannot fully observe directly and is relying on the insured's own account of its operations to do so accurately.

An AI agent that can commit a business to customer-facing statements, initiate transactions, or make decisions that affect third parties is precisely the kind of operational fact a prudent underwriter would want to know about before pricing a professional liability, cyber, or general liability policy. The reasoning is the same as it would be for any other material change in how a business operates: a business that hires a new class of decision-maker, human or automated, with the authority to bind it to consequential outcomes, has changed its risk profile in a way the insurer is entitled to know about.

The United Kingdom: the duty of fair presentation

The United Kingdom offers the clearest statutory example of how this duty operates today. The Insurance Act 2015 reformed the law governing non-consumer insurance contracts, replacing the historic duty of utmost good faith with a statutory duty of fair presentation. Under the Act, the insured must disclose every material circumstance it knows or ought to know, or failing that, provide the insurer with sufficient information to put a prudent underwriter on notice that further enquiries are needed.

Critically, the Act also reformed the remedy for breach. Before 2015, a material non-disclosure could allow an insurer to avoid a commercial policy entirely, regardless of how significant the specific omission was to the loss that actually occurred. Part 4 and Schedule 1 of the Act replaced this with a proportionate remedies regime. If the insurer would not have written the risk at all had it known the undisclosed fact, it may avoid the policy. If it would have written the risk on different terms, those terms apply retrospectively. If it would have charged a higher premium, the claim is reduced proportionately to reflect the premium that should have been charged. This is materially fairer to policyholders than the old regime, but it does not remove the underlying obligation, and an AI agent capable of consequential action is squarely the kind of fact this duty is meant to capture.

How EU jurisdictions differ

There is no single EU-wide insurance contract law harmonising disclosure duties across member states. Most EU civil law jurisdictions impose their own version of a pre-contractual disclosure duty under national insurance contract statutes, and the remedy for breach varies by jurisdiction and by the insured's state of mind. Some jurisdictions distinguish sharply between innocent non-disclosure, which may simply adjust the premium or terms, and deliberate or grossly negligent non-disclosure, which can void the policy entirely from inception.

The practical implication for an enterprise operating across multiple EU markets is that it cannot assume a single disclosure standard applies to every policy it holds. A group insurance programme spanning several EU jurisdictions, or a single policy governed by the law of a member state other than the insured's headquarters, may apply a different disclosure standard and a different remedy for breach than the enterprise's home jurisdiction would. Confirming the governing law of each policy, and the disclosure standard that law applies, is a necessary step before assuming AI agent disclosure obligations are uniform across a European group's coverage.

Underwriters are already asking directly

Any residual ambiguity about whether AI agent use must be volunteered is being closed by underwriting practice itself. Specialist AI liability carriers, including Armilla and Lloyd's market participants writing AI-specific endorsements, now ask about AI system deployment directly as part of the proposal or governance questionnaire, covering what the system does, what oversight exists, and what its scope of authorised action is, detailed in what AI insurance underwriters ask before writing a policy on this site. Where a proposal form asks a direct question about AI system use and the applicant answers incompletely or not at all, that is no longer a case of an undisclosed fact the insurer might have wanted to know. It is a direct misrepresentation on a specific question asked, which is treated more severely under disclosure law in most jurisdictions than a general failure to volunteer information the insurer never asked about.

Even on policies that do not yet ask about AI directly, such as legacy Commercial General Liability or Errors and Omissions policies renewed without an updated proposal form, the underlying material fact duty still applies. The absence of a specific question does not remove the general obligation to disclose facts a prudent underwriter would consider material.

What good disclosure actually looks like

Disclosure is not simply mentioning that AI is used somewhere in the business. A defensible disclosure describes, for each AI agent material to the risk being insured, what the agent does, what decisions or actions it can take without human review, what oversight and monitoring exist, and what the agent's track record has been in production, including any prior incidents. This is substantively the same evidence file that specialist AI underwriters request during their assessment process, described in the preparing an AI agent underwriting submission guide on this site, which means building it for disclosure purposes and building it for underwriting purposes is the same exercise.

A structured governance record, of the kind produced by an assessment against the Agent Certified framework, gives an enterprise a ready-made, evidence-backed disclosure statement rather than a hastily assembled account produced under renewal deadline pressure. It also creates a contemporaneous record that can later demonstrate the disclosure was made in good faith, which is directly relevant to how a non-disclosure dispute, if one arises, is likely to be resolved.

The cost of getting this wrong

The British Columbia Civil Resolution Tribunal's ruling in Moffatt v Air Canada confirmed that a business is bound by what its AI agent tells a customer, rejecting the argument that a chatbot operates as a separate legal entity.^[1] That liability exposure is exactly the kind of fact an insurer needs to know about before pricing cover. An enterprise that fails to disclose a customer-facing AI agent, experiences a Moffatt-type incident, and then seeks to claim under a policy that never priced that exposure, is asking an insurer to respond to a risk it never knowingly accepted. In the jurisdictions with the strictest non-disclosure remedies, the result can be a policy avoided from inception, meaning no cover at all, at precisely the moment the business needs it most.

Frequently asked questions