Does AI insurance cover EU AI Act fines?

A European enterprise facing a plausible Article 99 exposure under the EU AI Act, up to EUR 35 million or 7 percent of global annual turnover for the most serious violations, will reasonably ask whether an AI liability policy can pay that fine if the worst happens. The honest answer starts before any policy wording: in most of Europe, insuring a party against its own regulatory fine runs into a public policy restriction that predates AI entirely, and that restriction is unlikely to bend for Article 99. This article sets out why fines and third-party losses are treated so differently, what the closest working precedent, GDPR Article 83, actually shows, and what an AI liability policy realistically can and cannot do for you here.

Why "does insurance cover the fine" is really two questions

Ask an insurance question about a regulatory fine and it is tempting to treat it as a matter of finding the right policy with sufficiently broad wording. For AI Act fines, that framing skips the more fundamental question entirely. Before any policy wording is examined, most European jurisdictions apply a public policy doctrine that restricts or prohibits insuring a party against its own fine or penalty, on the reasoning that a punitive or deterrent sanction loses its function if the wrongdoer can simply transfer the cost to an insurer rather than bear it directly. Where this doctrine applies, no policy wording, however broad, can lawfully make the fine insurable, because the restriction operates at the level of insurance contract law and public policy, not at the level of what a specific insurer chose to underwrite.

This means the useful version of the question splits in two. First: is this category of loss, an administrative fine under Article 99, insurable at all in the relevant jurisdiction. Second, and only if the answer to the first is yes or partially yes: does a specific policy's wording actually extend to it. Most of the practical confusion in this space comes from skipping straight to the second question.

What Article 99 actually exposes an operator to

Article 99 of Regulation (EU) 2024/1689 sets a tiered penalty structure. Non-compliance with the prohibited AI practices listed in Article 5, including social scoring by public authorities and manipulative techniques causing harm, carries fines of up to EUR 35 million or 7 percent of the offending undertaking's total worldwide annual turnover for the preceding financial year, whichever is higher. Non-compliance with most other high-risk system obligations, including the Article 9 risk management system, Article 10 data governance, and Article 26 deployer duties, carries fines of up to EUR 15 million or 3 percent of worldwide turnover. Supplying incorrect, incomplete, or misleading information to notified bodies or national authorities carries a lower ceiling, up to EUR 7.5 million or 1 percent of turnover. In every tier, the higher of the fixed euro figure and the turnover-based percentage applies, which is why the largest AI Act fines are expected to be assessed against global turnover rather than the fixed ceiling for any company of meaningful size.

This tiered, turnover-linked structure is not a novel invention. It closely mirrors the penalty architecture the EU had already built for GDPR under Article 83, and European supervisory authorities and legal commentary have consistently drawn the comparison. That resemblance is exactly why the GDPR fine insurability experience is the most useful precedent available for thinking through Article 99, rather than treating AI Act fines as an entirely unprecedented category.

The GDPR precedent: what actually happened with Article 83 fines

When the GDPR's turnover-based fine regime under Article 83 took effect in May 2018, cyber insurers faced the same question enterprises are now asking about the EU AI Act: can a policy pay the fine itself. The answer that emerged over the following several years was jurisdiction-specific rather than uniform. In a number of EU Member States, courts and insurance regulators concluded, consistent with a long-standing principle across European insurance law, that insuring a company against its own administrative fine undermines the fine's deterrent purpose and is therefore restricted or void as against public policy, particularly where the underlying conduct involved an element of intent or serious negligence rather than pure strict liability. The practical market outcome was that GDPR fines are treated as largely uninsurable across most of the EU, while insurers continued, and continue, to offer cover for the legal defence costs, regulatory correspondence, and investigation expenses incurred while a GDPR enforcement action is contested, a materially different category of loss from the fine itself.

Insurers and brokers active in the AI liability space are, as of 2026, applying essentially the same analytical framework to Article 99. The turnover-based structure is comparable, the underlying public policy concern about undermining deterrence is identical, and no EU Member State has yet published guidance or case law suggesting AI Act fines will be treated more permissively than GDPR fines were. Operators should read the absence of an AI Act-specific fine insurability ruling as an open legal question rather than an invitation to assume coverage exists, not as a signal that AI Act fines will turn out to be easier to insure.

What AI liability insurers actually built instead

It is worth being precise about what the AI-specific insurance products active in 2026 are actually built to do, because none of them are marketed as fine-payment products, and that is not an oversight. AIUC-backed policies, following the AIUC-1 standard, cover hallucination-driven loss, data leakage, IP infringement, harmful outputs, and faulty tool actions, all framed as third-party losses the deployed AI system causes to a customer, counterparty, or affected individual. Armilla's Lloyd's-backed cover, with limits up to USD 25 million per organisation, covers performance errors, hallucinations, data leakage, and regulatory violations in a similar third-party liability frame, while explicitly excluding medical diagnostics and mental health applications. Munich Re's aiSure product settles on measurable performance data against an agreed specification, a structure built around performance shortfall, not around the operator's own regulatory sanction. Counterpart's Affirmative AI Coverage, triggered by hallucinations, misclassification, and deepfake fraud, sits within Miscellaneous Professional Liability and Tech Errors and Omissions insuring agreements, again a third-party liability frame.

The consistent pattern across every major AI-specific insurer active in the market is that the insured event is harm caused to someone else by the AI system, not a penalty imposed on the operator for how the system was governed or deployed. This is not a gap these insurers overlooked. It follows directly from the same public policy restriction discussed above: building a product around paying an operator's own fine would very likely be unenforceable in the jurisdictions that most matter to a European book of business, so no rational underwriter designs a flagship product around it.

The part that is realistically insurable: defence, not the fine

Where the GDPR experience is most directly useful is in showing what remains insurable even where the fine itself is not. Legal defence costs, meaning the fees for legal representation, expert evidence, and formal correspondence with a supervisory authority while an enforcement action or investigation is contested, are treated as a distinct category of loss from the fine in most jurisdictions, and are commonly available as an extension to professional indemnity, cyber, or directors' and officers' liability policies. This distinction matters practically because a serious Article 99 investigation can generate substantial legal costs well before, and regardless of whether, a fine is ultimately imposed, and those costs are the more realistic target for insurance planning.

The practical question to put to a broker is therefore not "does my policy cover an AI Act fine," which in most of Europe the law itself is likely to answer before the policy wording does, but "does my policy include a regulatory investigation or defence cost extension, and does it name AI Act enforcement as a triggering event." A policy that answers yes to the second question is doing real, usable work for a business facing Article 99 exposure, even though it cannot lawfully pay the fine at the end of the process. For the broader documentation posture that supports this kind of conversation with an underwriter, see our companion analysis of how AI compliance documentation functions as an insurance evidence chain.

Why this varies by Member State, and what that means for a multinational operator

The EU AI Act itself is silent on the insurability of Article 99 fines. Insurability of fines and penalties is a matter of national insurance contract law and public policy doctrine, which the Regulation does not harmonise. Member States differ both in how strictly they apply the doctrine against insuring deliberate or quasi-criminal penalties, and in how they characterise an administrative fine of this kind along the spectrum between an ordinary civil penalty and a punitive sanction. A business operating in several Member States should expect the answer to differ by jurisdiction rather than assume a single European position, in the same way that GDPR fine insurability case law developed unevenly across the bloc rather than settling on one EU-wide rule. Practically, this means a genuinely pan-European AI deployer's compliance and insurance planning should be built around the most restrictive plausible jurisdiction in its operating footprint, not the most permissive one, since the cost of assuming coverage that later turns out to be void is higher than the cost of planning conservatively. For the underlying obligations that determine whether an Article 99 fine is even in play, see the Article 99 penalties and fines deployer guide on agentliability.eu, and for how the revised Product Liability Directive adds a separate strict liability exposure layered on top of any regulatory fine question, see our analysis of Product Liability Directive coverage readiness.

What to check before you assume a fine is covered

Three checks are proportionate for any business assessing this exposure. First, ask your broker directly, in writing, whether the policy's fines and penalties exclusion addresses Article 99 specifically, and in which jurisdiction's law the policy is written to determine insurability, since that governing law question is often decisive. Second, confirm whether a regulatory investigation or defence cost extension exists separately from the fines and penalties question, since this is the coverage most likely to actually respond. Third, treat the compliance documentation your business holds under Articles 9 through 17 and Article 26 as the primary risk reduction tool here, not the insurance policy, since preventing the fine from arising is the only fully reliable answer available while the insurability question remains unresolved in most Member States.