Skim time: 6 minutes · Listening time: 6 minutes. Prefer to listen? Flip reads this week’s Brief here.

---

Two numbers this week.

The first is $205 million: the charge Markel took at State National, a major U.S. fronting carrier, after a bankrupt backer’s collateral proved, by Markel’s account, real but too small.

The second is $5,790: a premium an AI model supplied for an insurance application whose premium field was blank. It cited a page for it. The figure wasn’t in the document.

One number was set years ago against a reserve estimate that kept moving. The other came from a model that filled a blank with a guess. Both get tested the same way: years later, when a claim moves and someone asks where the number came from.

If you run a managing general agent (MGA), your renewal terms may already be moving, as may your own errors and omissions (E&O) and directors and officers (D&O) cover.

• Why 272 AI experts worry most about how fast firms adopt it.
• Collateral that was real, sized to a number that kept moving.
• An AI model, a blank premium field, and a page citation for a figure that wasn’t there.

---

272 experts put finance in AI’s top three

The hardest AI risk to see is a pressure.

MIT FutureTech and the University of Queensland asked 272 AI experts to rate 24 AI risks through 2030. In their judgment, information, national security and finance are the most vulnerable sectors. It’s expert opinion, not loss data.

In finance, they pointed to fraud, cyber, market manipulation, privacy breaches and failures in systems with wider economic reach.

The finding I’d underline is about competition. When firms believe speed is the advantage, they move fast, resist constraints and underinvest in safety. The researchers call it “an instrumental risk that creates other risks.”

The people most exposed to AI risk, its users and the people it affects, aren’t the ones with primary responsibility for managing it.

(source: MIT Sloan, July 20, 2026, on research by MIT FutureTech and the University of Queensland)

So what?

That mismatch is a governance problem, and governance is where D&O lives: the cover that can respond when a board’s oversight is challenged, subject to the policy’s terms.

An MGA board is better placed if its minutes show it asked what the tools do, who checks the output, and what happens when the output is wrong. What the board approved also has to match what the company tells its insurers about AI on the renewal application. A gap there can give an insurer grounds to contest coverage, or in some cases to seek rescission.

The next story has no AI in it, just the same habit in a fast-growing market.

---

Markel’s $205M fronting loss and your MGA’s capacity

Our read: long-tail programs should expect a collateral question.

The call. A fronting carrier is the licensed insurer issuing an MGA’s policies. It may retain some risk, transfer the balance to reinsurers and other capital providers, and hold collateral against what those backers owe. Markel said it had recognized a $205 million reserve at State National, which its CFO called a bad-debt charge: the portion of what a bankrupt backer owes that its collateral won’t cover. Five programs written from 2012 through 2021, mostly primary habitational casualty, were backed by that one capacity provider, which Markel hasn’t named. The claims kept developing after Markel stopped writing the business.

I read the transcript twice. An analyst asked whether the collateral was missing, bad, or too small. CEO Tom Gayner: “It’s the latter. The collateral is fine.” The losses, he said, “have moved at such a rate that it got ahead of the collateral.”

Where it broke. Vesttoo made the market ask whether collateral was real. State National highlights a different failure mode: valid collateral can still be inadequate when losses develop beyond the reserve estimate used to set it. On long-tail casualty, that reserve estimate is often the weakest number in the file.

I learned to read reinsurance at JLT, beside the old Towers reinsurance team, and what stayed with me was how much of a program’s safety comes down to a number somebody agreed on in a spreadsheet, years before anyone knows the answer. For your program, the class may be contractors, trucking, or claims-made professional lines. Claims-made business develops faster, but it still carries years of unreported claims.

What held. State National issued the policies, so State National owes the claims, and Markel absorbed the shortfall. Markel’s Andrew Crowley said the work “will result in some collateral top-ups for similar lines of business with other reinsurers who are in financially healthy positions.” Read that as an MGA: healthy counterparties may still get a collateral call when reserve estimates rise.

Separately, for its own U.S. casualty book, Markel’s Simon Wilson said a typical $10 million line is now closer to $5 million.

(source: Markel Q2 2026 earnings call, July 30, 2026, The Motley Fool transcript; Coverager, July 31, 2026)

The LION Lens

What happened — Markel took a $205 million bad-debt charge after a bankrupt backer’s collateral fell short on five habitational casualty programs (Markel Q2 2026 earnings call).

Why it matters — The collateral followed a reserve estimate that kept rising, on business that stopped writing in 2021.

Practical implications — Expect collateral, commission, line-size and data-access questions on long-tail programs, including ones you no longer write.

So what?

Every additional dollar of collateral a reinsurer posts is capital it can’t put to work elsewhere.

It may seek to recover that cost through lower ceding commission, tighter pricing, a smaller line, or a quiet exit. That’s more likely to appear in your next set of renewal terms than in a letter about State National. Unlicensed reinsurers and investor-backed funds can feel the pressure most, because collateral requirements may rise with ceded reserves.

It’s a model problem, not a people problem. Collateral set against today’s reserve estimate holds until the estimate moves, and on a long-tail book the estimate is likely to move.

Monday morning question: if your largest capacity provider failed tomorrow, would your front hold enough collateral to cover what it owes on your program, including claims nobody has reported yet? If you’d have to ask your front to find out, your front may already be asking the same question about you.

The LION POV

Here’s how we’re advising clients. Before your next renewal, check three things:

• Who stands behind your policies. Good is a capacity panel where no single provider’s failure would force your front to re-underwrite the program, plus notice rights if a provider is downgraded. The tell: one reinsurer or one fund carries most of your book.
• How the collateral is sized. Good is collateral set against unpaid ultimate losses, including claims not yet reported, with a margin and a named owner for the estimate. If the front placed the capacity, check the audit and information rights in your program agreement, and who pays for a top-up. The tell: collateral tied to reported losses and reviewed once a year.
• What your data lets them see. Good is claim-level loss data your front can reserve from without asking. The tell: summary reports only.

The same review belongs in your own E&O and D&O renewal, starting with notice, because it has a deadline. A front’s collateral review isn’t automatically a circumstance about the MGA. A review of your underwriting, binding, bordereaux, premium reporting, or delegated claims work may be. If it suggests a potential wrongful-act allegation, ask your broker and coverage counsel whether to give notice under your expiring claims-made E&O before it expires.

In the wordings we place, a front’s claim alleging negligent delegated work typically falls within the MGA E&O insuring agreement. Across the wider market, many forms carry a carve-out or sublimit for claims from a front or carrier, and the result still turns on carrier-client, contractual-liability, underwriting-loss, insolvency, and other wording, plus the retention. A demand resting only on a program-agreement indemnity isn’t the same as an allegation of negligent professional services, and a collateral top-up is generally a contractual funding obligation, not automatically E&O damages. A reserve charge can also bring shareholder allegations, including at a private MGA with outside investors. Ask whether your E&O covers regulatory proceedings and premium mishandling, and whether this renewal adds AI-related exclusions.

Read LION’s own MGA form the same way. It’s a Lloyd’s manuscript whose insuring agreement is written to address claims by the fronting carrier alleging a wrongful act in delegated work. It still has a retention, exclusions, and conditions, and the contract-versus-negligence point applies to it too.

If you’d rather walk into renewal knowing what your front will see, grab 30 minutes with me and bring your capacity panel and your E&O policy.

---

Your front wants claim-level data. Who’s reading it?

The third check above is data, and more of it is now read by AI first.

On the same July 30 call, Wilson said Markel has rewired six classes of business representing more than $500 million of premium with AI, cutting initial risk-assessment time by 50% to 90% by Markel’s measure.

On September 16, Cooper Labs published a benchmark testing 17 AI models on 166 insurance documents. On loss-run and statement-of-values questions, scores ranged from 61% to 86%. When a field was blank, redacted, or unreadable, models supplied unsupported values 10.6% to 38.3% of the time, depending on the model.

Cooper states important limits: it sells the software being tested, the documents are private, a single run’s score carries a 95% confidence interval of about six points, and it didn’t test Markel’s system. One model read an ACORD 125 (the standard commercial application) with the named insured redacted and the premium blank, then supplied both a company name and a $5,790 premium, with page citations.

Neither was in the document.

(source: Cooper Labs, Insurance Agent Benchmark, September 16, 2026; MarketBeat via TradingView on Markel’s AI rollout)

So what?

The speed on intake is real, and so is the risk that a plausible wrong number enters a loss file.

If AI touches your loss runs or bordereaux, the figures your front reserves from can inherit its mistakes. When a reserve moves in year six, the first question will be where the number came from. For an MGA, that can become an E&O question, depending on the wording and on who carries the loss under the AI vendor’s contract.

Monday morning action: ask which loss figures sent to your front this year were extracted by AI, and whether critical fields were checked in full or sampled. Confirm cited text appears on the cited page, reconcile paid plus reserves to incurred, log the model version, and require the tool to return “not found” rather than guess.

---

What LION is seeing in financial institution (FI) lines

On the renewals we’re seeing, MGA D&O is running about −3% to +1%, and MGA E&O 0% to +4%. These ranges come from renewals quoted from May through July, before Markel’s call.

For an MGA, rate is the smaller story. In our own portfolio this year, MGA E&O retentions have run from $10,000 to $100,000 for MGAs writing $5 million to $100 million of premium. None of those forms carried an express sublimit or carve-out for claims from the MGA’s own front. Outside our portfolio, many forms do. Even without an express front-claim limitation, the contract and notice points above still apply.

(source: LION book observations, renewals quoted May–July 2026; indicative, not quotes)

---

The Bottom Line

MIT’s experts warn that competitive pressure can move firms faster than their controls. State National’s collateral followed a reserve estimate that kept moving. The AI models in Cooper’s benchmark supplied values for blank fields and cited pages while they did it.

The number influencing your next renewal may have been set before the claim, and may already be stale. Check it before your front does.

Three for your board on Monday

1. Name the backer. Which capacity provider carries the most of our program? If it failed, would our policies still be honored, and who would absorb the shortfall?
2. Size the collateral. Do we have the right to see the security our front holds behind our business, and how does it compare with everything our backers could owe, including claims not yet filed?
3. Test the cover. If our own front brought a claim against us, would our E&O respond or carve it out? Is the limit sized to what that indemnity could cost? Have we given notice of anything we already know about?

A board can choose to retain a risk. It shouldn’t learn about one from its front’s collateral call.

If the honest answer to any of those is “I’d have to ask,” that’s the conversation to have before renewal. Reach out for a confidential conversation: book a time or reply to this email.

---

ICYMI: Last week closed The Endorsement Audit with a one-page renewal review covering all three parts: The Endorsement Audit, Part 3.

Want the D&O Contract Vigilance Blueprint, the five-day review of common D&O program gaps? Reply “blueprint” and I’ll send it.

Stay Covered Everybody,

-FLIP

P.S. Forward this to the person who owns your capacity relationships and the person who signs off on your loss data. Want to share it? Copy the link below:

https://lionspecialty.kit.com/posts/your-front-is-re-running-the-math

And if this was forwarded to you, subscribe here.

P.P.S. Nothing in this briefing constitutes legal advice. These are the opinions of the founder, and coverage is always subject to the terms of each policy. It’s market intelligence designed to help you ask better questions of your advisors and make sharper decisions at your next insurance renewal.

---

You're receiving this because you subscribed to the LION Specialty Boardroom Brief.
Unsubscribe  ·  Update your preferences