Who Insures the Machine

The Counterweight

Three essays in this series described a closing door. One traced how the industry converted its old silence about AI into printed exclusions, the way it had once done with cyber. One walked the specific ISO endorsements, CG 40 47 and CG 35 08, that took effect on the first of January and wrote generative AI out of standard general liability. One explained why the risk sits so far outside the actuary's tables that a careful underwriter would rather decline it than guess at its price. Read together, they leave a reader in a bleak place, holding a policy that no longer answers for the one exposure that keeps a modern operator awake. This essay is the counterweight. The door that closed on one side of the market opened a door on the other, and a real, funded, fast-moving industry has walked through it to sell exactly the coverage the exclusions took away.

The thing worth understanding at the outset is that this is not a promise or a pilot or a press release waiting for capital that never arrives. It is a market with names, capacity, and paying customers. Lloyd's syndicates have committed real limits to it. Munich Re, the largest reinsurer on earth, stands behind pieces of it. Managing general agents built specifically to underwrite machine risk are placing policies through licensed brokers today. A buyer who was told last quarter that AI had been excluded from the renewal can, this quarter, go into the market and purchase an affirmative grant of AI cover by its proper name. The point of what follows is to show who writes that cover, how the market that produces it is actually assembled, and why its architecture looks the way it does. The specific mechanics of how these policies pay, and the questions an underwriter will ask before issuing one, belong to their own essays and are treated there. Here the subject is the market itself.

There is a useful irony in where the answer landed. The same marketplace that pioneered the pricing of unnamed maritime perils three centuries ago, and that issued the 2019 mandate forcing cyber out of silence, is the marketplace now building the first structured home for AI liability. Lloyd's did not merely shut the door on silent AI. It convened, capitalized, and incubated the ventures that reopened it as priced, affirmative, and deliberately underwritten cover. The counterweight was engineered inside the very institution that dropped the first weight.

How the Market Is Built

To see who insures the machine, you first have to see the scaffolding, because affirmative AI cover is not being written by household-name carriers filing admitted forms in fifty states. It is being written on a different kind of frame, one designed precisely for risks that are new, hard to price, and potentially large. That frame has four parts worth naming plainly, because the rest of this essay leans on all of them.

The first part is the accelerator. The Lloyd's Lab is the market's InsurTech accelerator, a ten-week programme housed inside the Lloyd's building that takes start-ups and scale-ups and helps them develop a working insurance product for the Lloyd's market, with mentorship from experienced underwriters and direct access to capacity.^[1] Every six months it receives more than two hundred applications for roughly ten places, and its alumni have generated hundreds of millions in Lloyd's premium. When you read that an AI insurance venture is "Lloyd's Lab-backed," this is what the phrase points to. The venture was selected, coached, and connected to underwriting capital by the market itself, which is a meaningful signal about how seriously that market is taking the risk.

The second part is the managing general agent, and it is the structure doing most of the heavy lifting here. A managing general agent, an MGA, is an intermediary vested with actual underwriting authority delegated from an insurer, authorized to transact business and often to bind the insurer, issue policies, and arrange reinsurance under the terms of its agreement.^[2] At Lloyd's the equivalent role carries a specific name, the coverholder, defined by the market as a company authorized by a Lloyd's managing agent to enter into contracts of insurance on behalf of a Lloyd's syndicate under a binding authority agreement.^[3] The distinction matters because it explains how a two-year-old technology venture can issue a policy carrying the balance sheet of a centuries-old syndicate. The venture holds the expertise about the risk. The syndicate holds the capital. The coverholder agreement is the legal bridge that lets the first underwrite on behalf of the second.

The third part is the surplus lines market, and it is where almost all of this cover actually lives. Surplus lines, also called excess and surplus or E&S, is the non-admitted segment of the property and casualty industry, made up of specialist insurers that cover risks the standard admitted market will not write.^[4] The risks that land there share a profile: adverse or absent loss history, unusual characteristics, or a shortage of capacity in the standard market. That profile is a near-perfect description of AI liability, which has almost no loss history, behaves unlike anything on the actuarial record, and finds little appetite among admitted carriers. The surplus lines market exists to price exactly the risks nobody else will name a number on, which is why the affirmative AI policies described below are overwhelmingly surplus lines paper rather than admitted forms.

The fourth part is the reinsurer, standing behind all of it. Reinsurance is the transaction in which one party, the reinsurer, agrees for a premium to indemnify another party, the primary or ceding insurer, for part or all of the liability that insurer assumed under the policies it issued.^[5] In plain terms it is insurance for insurers, and it is what lets a primary carrier write a large or unfamiliar risk without betting its solvency on a single catastrophic year. When Munich Re or a comparable reinsurer stands behind an AI product, it is lending the balance-sheet depth that makes the primary cover credible. The correlated nature of AI risk, where thousands of deployers lean on the same handful of models, is precisely the kind of accumulation that a primary carrier wants a reinsurer to help absorb.

The venture holds the expertise about the risk. The syndicate holds the capital. The coverholder agreement is the bridge that lets the first underwrite on behalf of the second.How machine risk gets a balance sheet

Hold those four parts together and the shape of the market comes into focus. An accelerator inside Lloyd's incubates a specialist venture. That venture becomes a coverholder or an MGA with delegated authority. It writes affirmative AI cover as surplus lines paper. Syndicates and reinsurers stand behind the limits. Every affirmative AI product on the market today is some arrangement of those same components, and once you can see the frame, the individual carriers stop looking like a scattered list and start looking like variations on a single, coherent structure.

The Purpose-Built Underwriters

The clearest evidence that a real market has formed is the arrival of underwriters built for nothing else. These are not established carriers bolting an AI clause onto an old form. They are ventures whose entire reason to exist is to price machine failure, and their emergence is the surest sign that AI liability has crossed from a footnote into a class of business.

Testudo is the cleanest example of the full structure at work. It is a managing general agent, backed by the Lloyd's Lab, that writes generative AI liability cover for the companies that deploy and build AI. Its policies answer for third-party claims arising from AI-generated outputs, including hallucinations, model drift, and deviations from expected behavior, and they extend to IP infringement, unauthorized data disclosure, bodily injury and property damage, and regulatory proceedings, in the very scenarios where a general liability policy may now decline to respond. In early 2026 Testudo expanded its capacity to $9.25 million per insured by adding the Lloyd's syndicates Atrium and QBE to a panel that already included Apollo, aimed squarely at mid-market US enterprises.^[6] Read that sentence against the scaffolding: an MGA, incubated by the Lloyd's Lab, writing standalone AI liability on the capacity of named syndicates, with limits growing as more of those syndicates commit. It is the four-part frame made concrete in a single product.

Armilla built a comparable venture from the certification side. Its affirmative AI liability policy, launched in 2025 and underwritten by the Lloyd's insurer Chaucer, was among the first cover written to trigger on AI underperformance itself, on the hallucinations, drift, and critical errors that lead to damages, with legal defence and liability included.^[7] The important structural fact is the same one that appears everywhere in this market: a specialist venture supplying the AI expertise, a Lloyd's underwriter supplying the capital, and an affirmative grant of cover written by its proper name rather than left to the mercy of an unread exclusion. Armilla also offers a separate performance-warranty product that pays when an AI misses contractual accuracy or bias thresholds, and that instrument works on a genuinely different mechanism from a liability policy. That distinction between a guarantee and an indemnity is important enough to have its own treatment in this series, and it is not the subject here. What matters here is that Armilla exists at all, as one more purpose-built underwriter of machine risk sitting on Lloyd's capacity.

Relm approaches the same demand from Bermuda, the world's other great specialty market, and its lineup is instructive because it maps so neatly onto the gap the exclusions created. Relm launched a suite of AI liability solutions, and one of them, PONTAAI, is described by the carrier in its own words as an excess difference-in-conditions wrap designed to address the exclusions and gaps that now reside inside clients' existing liability programs as a result of their use or development of AI.^[8] A DIC wrap, recall, sits on an excess basis over an underlying program and fills what the underlying policies leave open.^[9] That makes PONTAAI perhaps the most literal market answer to the 2026 exclusion wave that exists: a policy whose entire job is to cover the AI liability that a cyber, professional indemnity, or general liability program now expressly declines. The exclusion opens a hole; the DIC wrap is engineered to plug precisely that hole, and nothing more.

The Incumbents Move In

A market made only of start-ups would be a curiosity. What makes this one durable is that the largest and most established names in insurance have decided AI cover is a business worth writing, and they are moving with the deliberateness that only a large balance sheet allows. When Munich Re commits, the signal is different in kind from a venture launch, because a reinsurer of that scale does not chase a fashion.

The most consequential incumbent move for the ordinary business arrived in March 2026, when HSB, a Munich Re company, introduced what it called the first standalone AI liability product built for small and mid-sized businesses. It answers for lawsuits arising from the use of AI, including bodily injury, property damage, and the advertising injury that flows from AI-generated advertising, marketing, blogs, and social media, and it was designed explicitly to fill the gaps that general liability policies now exclude.^[10] The distribution model is as telling as the cover. HSB does not sell the product directly to the business owner. It adds the cover to the policies of partner carriers, embedding AI liability inside programs those carriers already sell, subject to state regulatory approval. That is how coverage reaches the mass market. It does not arrive as a boutique standalone policy the small operator must go find; it arrives folded into the business insurance that operator already buys. This site treats the HSB product in its own dedicated explainer, and the launch marks the moment AI liability began its descent from the enterprise tier toward the corner shop.

CFC took a broader path to the same destination. Rather than build a single standalone AI policy, CFC embedded affirmative AI language across seven of its core products at once, spanning technology errors and omissions, professional liability, eHealth, intellectual property, management liability, media, and cyber, with the policies written to address AI-specific exposures such as hallucinations, AI-generated content, and model drift.^[11] The strategic choice is worth pausing on. Where a standalone policy says "here is a separate thing you may buy for AI," CFC's approach says "AI is now affirmatively addressed inside the coverages you already hold." Both are answers to the exclusion problem, and they represent the two distinct paths the market is taking: the dedicated standalone grant, and the affirmative clause woven back into the traditional lines that the exclusions had begun to strip.

Vouch shows a third path, the one aimed at the youngest companies. Vouch built its AI insurance for the startup segment through an integrated partnership with Corix, a division of Hiscox, giving VC-backed founders access to cover designed for the novel risks of AI, including errors and omissions for harmful or misleading model outputs, algorithmic bias, regulatory investigation defence, and IP infringement claims tied to training data or generative outputs.^[12] The Corix tie places a household-name carrier, Hiscox, behind a digital distributor built for founders who would never navigate a traditional surplus lines placement on their own. It is the same underlying structure, capital from an established carrier delivered through a specialist front end, tuned for a specific and underserved segment.

Read across these three, a pattern emerges that should reassure any buyer worried this market is a passing enthusiasm. The standalone specialists proved the risk could be underwritten at all. The incumbents, watching that proof, are now bringing scale, admitted-adjacent distribution, and reinsurance depth to it, each carving out the segment it knows best, HSB the small business, CFC the mid-market across many lines, Vouch and Hiscox the venture-backed startup. A risk that the largest reinsurer in the world is prepared to stand behind, across multiple distribution models at once, is not a fad. It is a class of business establishing itself.

Why Now, and How Fast

It is fair to ask why this all happened at once, and the answer is that the exclusion and the market are the same event seen from two sides. The instant the standard policy stopped covering AI by silent default, a priced and named demand appeared where a free and unnamed one had been. Every operator who lost silent AI cover at renewal became, in the same moment, a prospective buyer of affirmative AI cover. The exclusion did not merely remove protection; it manufactured a market by converting a diffuse, unbilled exposure into a specific thing a business now has to go out and purchase. The carriers described above are the supply that rose to meet a demand the exclusions created on the demand side.

The direction of travel is unmistakable even where precise figures for this young niche are still forming. The broader market for artificial intelligence inside insurance, which includes the tools carriers use as well as the cover they sell, has been projected to grow from roughly twenty billion dollars in 2025 toward well over a hundred billion by the early 2030s, a compound annual rate above thirty percent.^[13] That figure measures a wider category than AI liability cover alone, and it should be read as a directional signal about how central AI has become to the industry rather than as a precise size for this specific class. The narrower truth, visible in the product launches themselves, is that affirmative AI liability went from a single pioneering policy to a crowded field of specialists and incumbents inside a span of months, which is a pace of formation that insurance markets almost never manage.

What that pace means for a buyer is concrete. A business told at its last renewal that AI had been excluded is not, in fact, stranded. The cover exists, by name, from more than one source, structured through the most established institutions in the industry. What has changed is that the buyer now has to do the thing the era of silence never required, which is to go into the market deliberately, identify the product that fits the exposure, and purchase it as a named grant rather than lean on a default that no longer holds. The counterweight is real, but it does not attach itself. It has to be bought.

The natural next question, once a buyer knows the market exists, is which of these instruments actually fits, and that is a question of mechanism and of qualification rather than of names. Some of these products pay as liability policies, indemnifying you against third-party claims. Others operate as performance guarantees that pay on a measured failure without any allegation of fault. The difference decides what a policy is worth to you, and it has its own essay in this series. And before any of them will issue, an underwriter will ask a specific set of questions about how you govern the AI you deploy, which is the subject of another. The place to begin, though, is simply to see the field whole, and the comparison built for that purpose lays every one of these carriers side by side.