# Where to build the agent insurer: underwrite and carry the risk in London, sell into the US and the EU, raise the capital in one go, and what £10 million or £50 million buys in cover, sgit.ai

> The last three articles described insuring an agent row by row from its behaviour policy, who could carry the risk, and the vendor's share of it. This one asks where such a company should be built, from a regulatory and an investment point of view, and how much it would take. My answer is London first, for two of its three jobs. It should start as a managing agent there, pricing and selling on a Lloyd's syndicate's capital, because that is where two of today's AI agent covers are written, authorisation takes up to nine months, and an intermediary's own capital is the higher of £5,000 or 2.5% of its income; and it should carry the risk there too, as a UK insurer through the PRA's mobilisation regime or with capital at Lloyd's. It should sell into the US through surplus lines, a $143.3 billion market built for risks without a loss history, and into the EU by one registration, once Europe's product rules make software a product on 9 December 2026. The Gulf's funds are a place to raise capital. On the money, I built a small model with every assumption stated, and the point it makes is that what the money buys is cover. Two years of running the company cost about £4.35 million; everything raised above that backs policies, and because every row is capped, the ceiling of cover follows from it. Raised in one go, £10 million backs about £94 million to £188 million of cover, 1,177 to 2,354 agents at £80,000 each; £50 million backs about £761 million to £1.52 billion, 9,510 to 19,020 agents, depending on how much of the book one shared failure could hit at once. The bigger the cover the company aims for, the more it should raise at the start.

*Source: <https://newsroom.sgit.ai/articles/where-to-build-the-agent-insurer.html> · sgit.ai v0.7.87 · this file is generated from the same content as the page, so the two cannot drift. Every page on this site has a `.md` twin; internal links below point at them.*

---

[Home](../index.md) / [Articles](index.md) / Where to build the agent insurer: underwrite and carry the risk in London, sell into the US and the EU, raise the capital in one go, and what £10 million or £50 million buys in cover

# Where to build the agent insurer: underwrite and carry the risk in London, sell into the US and the EU, raise the capital in one go, and what £10 million or £50 million buys in cover

By [Dinis Cruz](https://sgit.ai/about/index.md) · 2026-10-28 · [article v1.2.0, 3 versions](versions/where-to-build-the-agent-insurer.md) · [newsroom v0.13.127](https://newsroom.sgit.ai/admin/versions.html) · insuranceunderwritingagent-behaviour-policyriskmandatecapitalregulationlloydsfundraisingstartupsinsuring-agentsarticle

***Abstract:** The last three articles described insuring an agent row by row from its behaviour policy, who could carry the risk, and the vendor's share of it. This one asks where such a company should be built, from a regulatory and an investment point of view, and how much it would take. My answer is London first, for two of its three jobs. It should start as a managing agent there, pricing and selling on a Lloyd's syndicate's capital, because that is where two of today's AI agent covers are written, authorisation takes up to nine months, and an intermediary's own capital is the higher of £5,000 or 2.5% of its income; and it should carry the risk there too, as a UK insurer through the PRA's mobilisation regime or with capital at Lloyd's. It should sell into the US through surplus lines, a $143.3 billion market built for risks without a loss history, and into the EU by one registration, once Europe's product rules make software a product on 9 December 2026. The Gulf's funds are a place to raise capital. On the money, I built a small model with every assumption stated, and the point it makes is that what the money buys is cover. Two years of running the company cost about £4.35 million; everything raised above that backs policies, and because every row is capped, the ceiling of cover follows from it. Raised in one go, £10 million backs about £94 million to £188 million of cover, 1,177 to 2,354 agents at £80,000 each; £50 million backs about £761 million to £1.52 billion, 9,510 to 19,020 agents, depending on how much of the book one shared failure could hit at once. The bigger the cover the company aims for, the more it should raise at the start.*

Underwrite and carry the risk in London, sell into the US and EU, raise the capital in one go.

Insuring agents

Five articles and three vaults on one idea: an agent's behaviour policy read as an insurance policy, priced row by row, split with the vendor, and backed by capital.

[1 · The definitionA payout for every rowA limit per row of the behaviour policy, a lower premium per enforced control, and no cover for what it does not name.](a-payout-for-every-row.md)[2 · The marketWho carries the riskWho sells AI agent cover today, the four parts of an insurer, and starting as a managing agent.](who-carries-the-risk.md)[3 · The splitThe vendor's layerA share of each row the agent's vendor carries, when its own conditions held.](the-vendors-layer.md)[you are here4 · Where and how muchWhere to build the agent insurerLondon to underwrite and carry the risk, and what £10 million or £50 million buys in cover.](where-to-build-the-agent-insurer.md)[5 · The whole storyInsuring agents, the whole storyThe four articles and the three vaults, in one place.](insuring-agents-the-whole-story.md)[Vault · posrhzp3Licence to OperateOne agent's policy, simulated: every reply priced against a band, a ceiling, a pool and a premium.](https://dev.vault.sgraph.ai/#sgit_public_read_d990a52efb9af32c8463e2962f3ca5ccf92b3b6e8ea788e55009073c29b4da29%3Aposrhzp3)[Vault · m1wbrefrThe underwriting calculatorThe series' formula as data, with a calculator and a script that reproduces every number.](https://dev.vault.sgraph.ai/#sgit_public_read_54a5d5d39863f124c6a304b20874e53436d77df42945ba15c74712a680b22991%3Am1wbrefr)[Vault · 2wzct4k7AIUC-1, as a graph you can citeThe standard as a graph, with a conformance layer that asks what would be insurable. Unofficial.](https://dev.vault.sgraph.ai/#sgit_public_read_0f01d367b04f886f6c65038649b76504f4cd2ad06480d88a5e933a671e0db072%3A2wzct4k7)

**What came next.** [Insuring agents, the whole story](insuring-agents-the-whole-story.md) ties the five articles together and shows the three vaults where the ideas run.

## In short

- **The company does three jobs, and London suits two of them.** Pricing and selling, and carrying the risk; distribution goes where the customers are.
- **Start as a managing agent in London.** Two of today's AI agent covers are written by Lloyd's coverholders; authorisation takes up to nine months; the intermediary's own capital is small.
- **Carry the risk in London too.** A UK insurer through the PRA's mobilisation regime, or capital at Lloyd's behind a syndicate.
- **Sell into the US through surplus lines.** A $143.3 billion market in 2025, built for risks no filed policy covers yet.
- **Reach the EU by one registration,** once software becomes a product there on 9 December 2026; EIOPA is wary of MGAs, so expect questions.
- **What the money buys is cover.** About £4.35 million runs the company for two years; everything raised above that backs policies.
- **£10 million in one go backs about £94 million to £188 million of cover.** 1,177 to 2,354 agents.
- **£50 million in one go backs about £761 million to £1.52 billion.** 9,510 to 19,020 agents; the bigger the aim, the more to raise at the start.

## The question, and why it has more than one answer

In this series I described [insuring an agent row by row](a-payout-for-every-row.md), [who could carry that risk](who-carries-the-risk.md), and [the vendor's share of it](the-vendors-layer.md). The second article ended with a path: start as a managing agent on someone else's capital, prove the loss record, then carry the risk. So the natural next question is where to build such a company, from a regulatory and an investment point of view: the US, the UK, the EU, or the Middle East, where there are new regimes and new money.

I think the honest answer is that it can have more than one home, because it does more than one job. A managing agent prices and sells policies on an insurer's behalf and carries no risk on its own balance sheet; it needs a licence to distribute insurance and a carrier willing to lend its capital. An insurer pays the claims and has to hold capital against a bad year. Distribution needs licences where the customers are. Those three jobs are regulated differently in each place, and the best place for one is not the best place for the others.

The UK, US, EU and UAE: what each asks of a managing agent and an insurer.

## Start in London

The first job is pricing and selling, and London is where I would start. The AI agent cover on the market today is written in a handful of ways, and two of them, Armilla and Testudo, are Lloyd's coverholders with Lloyd's syndicates behind them, as [who carries the risk](who-carries-the-risk.md#landscape) set out. A coverholder needs a Lloyd's broker or managing agent to sponsor it, and the market is used to new specialty risks: the Lloyd's Lab's current cohort includes a "Risk engine quantifying advanced and systemic AI risks" ([FinTech Global](https://fintech.global/2026/03/20/lloyds-lab-unveils-cohort-16-insurtech-innovators/), 20 March 2026).

The UK licence is within reach of a start-up. A managing agent is either authorised directly by the FCA, which "can typically take up to nine months from when an applicant starts preparing its application", or works as an appointed representative of a firm that is; its own capital requirement, if it does not hold client money, is the higher of "GBP5,000 or 2.5% of annual income from regulated insurance distribution activity", with professional indemnity cover on top ([DLA Piper](https://intelligence.dlapiper.com/managing-general-agents/?c=GB), September 2026). The same city has the regime for the next step, below.

## Sell into the US

The US is the largest market, and the way a new risk reaches it is surplus lines. They cover "unique risks for which admitted carriers do not offer a filed policy form or rate", and the industry's own association describes new products with no loss history as "difficult, if not impossible, to price or rate for insurance purposes" ([WSIA](https://www.wsia.org/wcm/wcm/About/What_is_Surplus_Lines.aspx)). That is exactly an agent behaviour policy today. The market is large and growing: "The surplus lines market closed 2025 at a record $143.3 billion in direct premium written" ([Insurance Business](https://www.insurancebusinessmag.com/us/news/excess-surplus/surplus-lines-hit-143bn-record-as-placement-market-shifts-589475.aspx), citing AM Best, 11 September 2026). Harbor's AI agent cover is issued by surplus lines carriers.

A US managing agent needs a producer licence in each state it writes in, and the NAIC's model law asks for a bond of "at least $100,000 or ten percent (10%) of the managing general agent's total annual written premium ... but in no event greater than $500,000" ([NAIC](https://content.naic.org/sites/default/files/model-law-225.pdf)). The US MGA market reached "approximately $128 billion in 2025", and "approximately 20% of total US MGA premium is supported through fronting carrier relationships" ([Reinsurance News](https://www.reinsurancene.ws/conning-estimates-us-mga-premiums-reached-128bn-in-2025/), citing Conning, 29 July 2026). So a London managing agent can reach US customers through the surplus lines market, with a US entity when the book justifies one.

The US also has the cheapest way for a provider to insure its own agents: a captive. In Vermont, a pure captive needs capital of "not less than $250,000.00" ([Vermont statutes](https://legislature.vermont.gov/statutes/section/08/141/06004)). The providers who, as [who carries the risk](who-carries-the-risk.md#providers) found, have looked at insuring their own risk could start there.

## The EU, second

The EU is the market where demand may arrive fastest, because from 9 December 2026, as [the vendor's layer](the-vendors-layer.md#europe) described, software is a product under the new product liability directive. Distribution is simple in principle: an intermediary registered in one member state works across the union on "the principle of a single registration" ([Central Bank of Ireland](https://edit.centralbank.ie/regulation/industry-market-sectors/brokers-retail-intermediaries/insurance-brokers-intermediaries/passporting-in-and-out)). An insurer's minimum capital under Solvency II starts at €2.7 million.

The supervisor is cautious about exactly this model. EIOPA wrote in June 2026 that "insurers are increasingly outsourcing key functions to MGAs, creating concerns about weak oversight, conflicts of interests, unclear responsibility for product governance and difficulties in cross-border supervision", and that it "could engage with the Lloyd's market to better understand how we can build capacity in the EU" ([EIOPA](https://www.eiopa.europa.eu/document/download/e4de2f84-0a0a-42ed-82b6-b8b5620c7745_en?filename=2026-06-17+-+EIOPA+Note+on+Competitiveness.pdf)). A company whose product is a written, deterministic policy with clear governance should answer those concerns better than most; I would still plan the EU as the second market, not the first home.

## Carry the risk in London too

When the company is ready to carry risk itself, I would do that in London as well, so that the underwriting and the balance sheet sit under the same supervisors and the same market. There are two ways in. The first is a UK insurer: the PRA has run a mobilisation regime for new insurers since 31 December 2024, "to facilitate new insurer expansion and entry into the UK market", an optional stage of up to twelve months with restrictions that begins at authorisation ([Bank of England](https://www.bankofengland.co.uk/prudential-regulation/new-insurer-start-up-unit)), and a general insurer's minimum capital floor is £2.4 million. The second is capital at Lloyd's: "A Lloyd's syndicate is formed by one or more members joining together to provide capital and accept insurance risks on a several liability basis", and the members' capital is held as Funds at Lloyd's ([Lloyd's](https://www.lloyds.com/market-resources/member-services/about/what-is-a-member)). A company that starts as a coverholder already works inside that market, so putting its own capital behind its own book is a short step.

An insurer that wants EU clients directly may need an EU carrier as well, under Solvency II's €2.7 million floor. That can wait for the second market.

## The Gulf, for capital

The Gulf has new regimes and a great deal of money, and I think its role here is the money. The UAE's two financial centres are both reworking their insurance rules: in Abu Dhabi, the ADGM's regulator proposed in December 2025 that insurers hold "120% of the ASCR under normal circumstances" ([Reinsurance News](https://www.reinsurancene.ws/fsra-proposes-comprehensive-updates-to-adgm-insurance-regulatory-framework/)); the DIFC's own pages could not be read, so its figures are left out. The region's funds invest in balance sheets of exactly the kind the third job needs. I would raise there, and leave selling there for later.

## How much it would take

You asked for an opinion on the investment, so here is mine, with the arithmetic behind it in [a data file](https://newsroom.sgit.ai/articles/data/where-to-build-the-agent-insurer__capital.json) and every assumption stated in the script that writes it.

**Running the company: about £4.35 million for two years.** A team of eight in the first year and fifteen in the second, at £150,000 each including everything, and £450,000 a year for regulatory, legal, actuarial, professional indemnity and technology costs: £1.65 million in the first year and £2.7 million in the second. The market is funding companies like this: insurtech funding was "US$5.08 billion" in 2025 ([Insurance Journal](https://www.insurancejournal.com/magazines/mag-features/2026/03/09/860638.htm), 9 March 2026, citing Gallagher Re), and in the first quarter of 2026 "InsurTechs linked to AI liability and cyber insurance attracted USD 444.84 million" ([Reinsurance News](https://www.reinsurancene.ws/global-insurtech-funding-holds-firm-as-ai-dominates-q126-investment-gallagher-re/), 7 May 2026). AIUC has raised $55 million in two rounds, and Harbor was reported this month to have raised a $10 million seed.

**Backing the policies: what the rest of the money buys is cover.** This is where the deterministic policy pays off. Because every row is capped and every unnamed capability excluded, the worst year can be bounded from the limits on the book: take all the limits sold, the share of them one shared failure could hit at once (one model's flaw, one tool's bug, every agent built on it), keep half after reinsurance, and add a 20% buffer. Turn that round, and every pound of capital raised above the running costs sets a ceiling on the cover the company can sell. So rather than raising in stages, the company can raise once, for the cover it intends to sell from the start, and the size of the raise decides that ceiling.

The ceiling of cover a raise of £10 million or £50 million can back. Illustrative, from capital.py.

With £80,000 of limits per agent, as in the worked example, and two years of running costs taken out first:

| Raised in one go | Capital backing policies | If 5% can be hit at once | If 10% can be hit at once |
|---|---|---|---|
| £10 million | £5.65 million | about £188 million of cover, 2,354 agents | about £94 million of cover, 1,177 agents |
| £50 million | £45.65 million | about £1.52 billion of cover, 19,020 agents | about £761 million of cover, 9,510 agents |

Both are above the UK's £2.4 million floor. At the worked example's rate, £6,720 a year for £80,000 of cover with five enforced controls, the £10 million book could earn about £7.9 million to £15.8 million of premium a year, and the £50 million book about £64 million to £128 million; those are ceilings at full use of the capital, not forecasts.

So my opinion: decide the cover first, and raise for it in one go. £10 million is enough to prove the model on about a thousand to two thousand agents; £50 million is the raise for a company that means to insure ten to twenty thousand agents from the start, and to be the place the large providers come to. The share hit at once is the number that matters most, and it is the one with no loss data behind it yet: it is why even a large book should start spread across models and tools.

## What this does not claim

- **This is not legal, tax or regulatory advice.** It maps what regulators publish; a real decision needs lawyers and regulators in each place.
- **The model's inputs are mine.** Team sizes, costs, limits, the shared-failure share and the reinsured half are assumptions chosen to show the shape; the capital a regulator would ask for comes from its own formula.
- **Currencies are approximate.** The floors are as each regulator states them; the model converts them to pounds roughly.
- **The Gulf figures are thinner.** The DIFC's pages could not be read; the ADGM's proposal is a discussion paper, not a rule.
- **The premium figures are ceilings.** They assume every pound of capital is used at the worked example's rate; a real book would sell less, and price each agent on its own rows.
- **Fronting fees are left out.** We found no reliable published figure for what a carrier charges to lend its licence and balance sheet.
- **Tax is left out.** Where to be taxed is a separate question from where to be regulated.

If you invest in insurers, or run one in any of these places, and would build this with us, write to [agent@riskmandate.ai](mailto:agent@riskmandate.ai).

## Where this comes from

A voice memo of mine on 28 October 2026, the third of this series, written by this site's agent in the voice of this site. The argument and the editorial responsibility are mine; the routes and the estimate are my opinion, as the memo asked.

Every outside source is linked where it is used. They were read on 28 October 2026 by a research agent of this site, which fetched each page quoted; where it could not, the fact is marked as secondary or left out. The research notes, with what could and could not be verified, are `admin/notes/2026-10-28__underwriting-abp/research-04.md` in this site's repository. The model is `admin/notes/2026-10-28__underwriting-abp/capital.py`, and its output the data file linked above.

On this site: [a payout for every row](a-payout-for-every-row.md), [who carries the risk](who-carries-the-risk.md) and [the vendor's layer](the-vendors-layer.md), the first three articles of the series.

## Threads

Startups & strategyAgents & policy[This article as a graph →](graphs.md#where-to-build-the-agent-insurer)

### Builds on

- [Insuring agents, the whole story: five articles and three vaults on one idea, that an agent's behaviour policy can be read as an insurance policy, priced row by row and backed by capital](insuring-agents-the-whole-story.md) The five articles on insuring agents in one place, and the three vaults where the ideas run: the simulation, the calculator and AIUC-1.
- [A payout for every row: underwriting an agent behaviour policy, where each risk it names has its own limit, each enforced control lowers the premium, and what it does not name is not covered](a-payout-for-every-row.md) Underwriting an agent behaviour policy row by row: a limit per named risk, a lower premium per enforced control, and no cover for what is not named.
- [Who carries the risk: the companies already insuring AI agents, the parts an insurer for agent behaviour policies would need, and why the first one starts as a managing agent on someone else's capital](who-carries-the-risk.md) Who already insures AI agents, the four parts an insurer for behaviour policies needs, why capital matters, and why the first one starts as a managing agent.
- [The vendor's layer: what the companies that sell AI agents should cover themselves, row by row, when their contracts cap liability at a year's fees and Europe's new product rules treat software as a product](the-vendors-layer.md) Each row of an agent behaviour policy gets a vendor's share, carried when the vendor's conditions held; contracts cap at a year's fees, Europe moves the line.

### Continued by

- [Insuring agents, the whole story: five articles and three vaults on one idea, that an agent's behaviour policy can be read as an insurance policy, priced row by row and backed by capital](insuring-agents-the-whole-story.md) The five articles on insuring agents in one place, and the three vaults where the ideas run: the simulation, the calculator and AIUC-1.
- [The vendor's layer: what the companies that sell AI agents should cover themselves, row by row, when their contracts cap liability at a year's fees and Europe's new product rules treat software as a product](the-vendors-layer.md) Each row of an agent behaviour policy gets a vendor's share, carried when the vendor's conditions held; contracts cap at a year's fees, Europe moves the line.

[All articles](index.md) · [All graphs](graphs.md)

Article No. 319 · [start a reading list with it](https://newsroom.sgit.ai/for/make.html#r=319)

**Posting this article on LinkedIn?** The cover is [where-to-build-the-agent-insurer.jpg](banners/where-to-build-the-agent-insurer.jpg) (1920×1080, title and key ideas on it). Upload it as the article cover, paste the title, then select and copy the body from this page.

**Want the next issue by email.** One issue a week or so: what was published, what it adds up to, and what is worth your time. [Subscribe to the SGit Newsroom →](../subscribe/index.md)

[← All articles](index.md)


---

*[Site index for agents](../llms.txt) · [HTML version](https://newsroom.sgit.ai/articles/where-to-build-the-agent-insurer.html)*