See how Forma Health supervises their AI

The One Question Every Insurance CEO Should Ask Their Board Before the Next Cat Season

AI GovernanceLast updated on
The One Question Every Insurance CEO Should Ask Their Board Before the Next Cat Season

If our AI drifts and our reinsurer walks, are we still a company?

That is the question. Most boards have not yet asked it explicitly, and the carriers that survive the next decade of AI-era underwriting will be the ones that can answer it before they are asked. We have spent thirteen posts on the supporting argument: the treaty mechanics, the rating agency methodology, the regulatory floor, the case law, the loss ratio dynamics, and the historical examples of carriers that closed inside ninety days of a reinsurance event. This piece collapses the argument into one CEO-level question, walks the five-part chain that produces the failure mode it is pointed at, and sketches the board agenda that puts a defensible answer on the table before the next June 1.

The audience is narrow on purpose. This is written for the CEO and the board members of a regional or specialty P&C carrier with a meaningful catastrophe XL program and an AI footprint somewhere in rating, underwriting, claims, or marketing. If both of those things are true at your company, the chain below is mechanically live against your balance sheet. The work to answer the question is achievable inside one renewal cycle, and the documentation it produces is reusable across rating reviews, ORSA exercises, and treaty placements.

The Five-Part Chain in One Place

The failure mode that closes the carrier runs five steps from a routine ML retraining cycle to receivership. Each step is supported by existing law, existing treaty wordings, or existing rating agency practice, and none of the steps is hypothetical on its own. The unique claim of this body of work is that AI shortens the chain and makes the disclosure question harder to answer in time.

The chain is this. The carrier's machine learning rating model retrains on recent loss experience and shifts its risk weights. The shift is large enough that the prior actuarial framework would have classified it as a change in underwriting philosophy. The carrier's reinsurance panel is not formally advised, because internal governance does not yet treat retraining as a treaty-disclosable event. A catastrophe loss develops larger than the new model priced for, often in a secondary peril that the historical training window underweighted. The reinsurer reviews the cession, identifies the model change as a breach of the disclosure warranty we walked through in the post on the one treaty clause your AI strategy lives or dies on, and rescinds, non-renews, or significantly reprices the layer. A subsequent cat loss attaches at the carrier's net retention, the layer above is gone or repriced beyond what surplus can support, and the loss flows through to capital. The carrier is insolvent.

The historical evidence that "treaty loss closes carriers" is not theoretical. FedNat lost its catastrophe placement on June 30, 20221 and was insolvent eighty-nine days later, on September 27,2 with no triggering event in the window. The full sequence sits in the FedNat piece. Merced Property and Casualty was AM Best A- rated and one hundred and thirteen years old when the Camp Fire ignited on November 8, 2018,3 and was insolvent twenty-five days later, on December 3,4 an outcome we treated in the Merced post. UPC's Hurricane Ian losses developed from $660M5 to $1.5B6 through the fall of 2022, exhausted the program, and put the carrier into receivership on February 6, 2023.7 We pulled the three Florida failure modes together in the Florida postmortem.

State Farm General produced the cleanest contemporary example. The California homeowners book sat behind a parent-level Q3 2023 loss ratio that jumped roughly twenty-six points year-over-year, and within months the carrier had paused new business in California8 and announced the non-renewal of approximately 30,000 in-force policies.9 The triggering pressure was reinsurance affordability, not a regulator. We unpacked the swing and the treaty underwriter's reading of it in the seventeen-point loss ratio piece.

These four cases pre-date the AI era as the cause of failure, and they are useful precisely because they establish the mechanism. The remaining question is whether AI-driven misclassification produces enough rated-but-uncovered exposure to trip the same wires, faster, at carriers nobody currently has on a watchlist.

Three Structural Facts Every Board Should Internalize

The first structural fact is that the reinsurance treaty is a balance sheet event, while the regulatory fine is a P&L event. Compliance budgets across the P&C industry are sized for the second category. The largest US insurance regulatory penalties of the past decade hurt the affected carriers without closing any of them. A treaty walking after a single bad cat year is a different category of event entirely, and surplus does not absorb that hit on its own. We made this case in detail in the canonical post on why the reinsurance treaty is where AI risk becomes existential, and the asymmetry remains the most underpriced element of AI risk in board packets we see.

The second structural fact is that the mechanism that closes carriers (treaty exhaustion, treaty non-renewal, rating cascade) is older than AI. The Florida market between 2020 and 2023 ran the cleanest natural experiment in carrier mortality the US insurance industry has produced in fifty years, and AI was incidental to none of the failures that came out of it. The chain we have been describing in this body of work is the same chain UPC, FedNat, and the Demotech-downgraded carriers walked. AI changes the trigger, not the chain. That distinction is important because it means the documentation, the disclosure discipline, and the ORSA stress tests that protect against AI-era treaty loss are extensions of governance practices reinsurers and rating agencies already understand. The work is incremental against frameworks that already exist, and the absence of a published AI-era precedent does not buy the carrier any time, because the underlying mechanism is one the legal and rating systems have already adjudicated repeatedly in the pre-AI cases.

The third structural fact is that the reinsurer is already underwriting AI. Munich Re sells aiSure, a product that underwrites the performance of a customer's AI model directly.10 Swiss Re signed a Memorandum of Understanding with RIQ at Abu Dhabi Finance Week in December 2025, focused on AI-enabled capacity solutions across the UAE.11 Major reinsurers are building internal models for AI failure modes against their own balance sheets, and those models travel back into how they evaluate cedents. A cedent that says nothing about its AI governance still has a view formed about it on the reinsurer's side, and the view eventually expresses itself at a renewal as a price, a warranty tightening, or a non-renewal. The cedent's silence reads as a data point in itself, and rarely a flattering one.

The Board Agenda Before the Next Cat Season

The CEO question turns into a board agenda item with five concrete deliverables, each producible inside a single renewal cycle and none requiring consultants the carrier has not already engaged for adjacent work.

A model change log written for treaty underwriters. The artifact a reinsurance underwriter needs is not a data scientist's commit history. It is a document that states what changed in the model stack, when it changed, and how the change affects the carrier's expected loss curve by line and territory. The change log should be dated, authored by a named risk owner, and refreshed at a cadence that matches the model retraining schedule. The board's role is to require its production and to verify, at least annually, that the document exists and has been delivered to the lead reinsurer.

A treaty disclosure register. The register documents what was formally communicated to which reinsurer at which renewal, with copies of the underlying correspondence attached. Most carriers we observe cannot produce this register in 2026, because treaty placement has been managed as a brokerage event rather than a governance event. The register is the artifact that supports the cedent's defense in the litigation that follows a future treaty rescission. Building it after the fact is meaningfully harder than building it as renewals close. The board can require the register to be produced and maintained without taking a position on what specifically should be disclosed at any given renewal.

An ORSA exercise that models treaty rescission. The standard carrier ORSA already includes a treaty exhaustion scenario, where the program pays through its layers on a large event and surplus absorbs the remainder. The scenario most ORSA exercises do not include is rescission, where the layer disappears before the event because the reinsurer has identified a disclosure breach. The two outcomes have different recovery paths and different surplus implications, and a board that has only seen the exhaustion case has a partial view of the worst quarter the carrier could have. Including the rescission scenario in the next ORSA cycle is operationally inexpensive and answers a question the board should already want answered.

A rating-trigger response runbook. The runbook assumes that an AM Best Enterprise Risk Management concern letter is a possible quarter-end outcome rather than a remote tail event. We walked the methodology in the AM Best Comprehensive Adjustment piece, and the takeaway for the board is that the qualitative override exists and that 2023 was a sharpening year. AM Best issued 55 P&C downgrades against 35 upgrades that year, well above the 30 downgrades the year prior, and the pattern continues.12 Demotech's mid-July 2022 letters to roughly seventeen Florida carriers, which we covered in the Demotech template piece, provide the closest precedent for how a rating action can move from letter to treaty consequence to receivership inside one quarter.13 The runbook names the responsible executive, the disclosure cadence, the reinsurance panel notification protocol, and the contingency capital plan for the worst-case sequence.

Board acknowledgment that ML retraining is treaty-relevant. This last item is documentary. The board's model governance policy should state, in writing, that retraining a production rating model is a treaty-disclosable event. The carrier can then operationalize that statement against whatever cadence and threshold the CRO and the chief underwriting officer agree on. Without the statement at the policy level, the disclosure question lives inside the data science team, where it is unlikely to be answered in a way a coverage attorney would defend.

These five deliverables together are the operating definition of "we have done the work" against the CEO question. They are also, helpfully, the operating definition of what a reinsurer reads as above-baseline AI governance against the regulatory floor we discussed in the NAIC bulletin piece. The same documentation supports both audiences.

The Honest Framing

No published case has yet adjudicated the AI-drift to treaty-rescission chain. We are not claiming otherwise. The mechanics are in place, the legal doctrines exist, the reinsurer-side capability is being actively built, and the litigated case has not been written. The Second Circuit's decision in Utica Mutual v. Clearwater, 906 F.3d 12 (2d Cir. 2018), vacated the cedent's claim that the reinsurer was bound to follow its settlement absent a demonstrated contractual obligation to do so,14 which narrows the most plausible cedent-side argument that a friendly contractual ride-through would soften the impact of a future rescission. The Lemonade "AI Jim" episode, which we treated in the Lemonade case study post, shows what happens when the disclosure question moves from the renewal table into the public domain. Hippo's Q1 2023 273% loss ratio,15 walked in the Hippo piece, shows what an AI-era pricing model does when the climate distribution moves outside its training window. Aon documented that secondary perils accounted for 86% of global insured losses in 2023,16 which is exactly the category of loss most likely to embarrass an ML model trained on a too-recent window.

The honest framing is that the train is on the tracks. The first carriers to be hit by it will not have seen it coming, because the entry point is governance documentation rather than a market event a CFO would be tracking on a quarterly dashboard. The work that prevents the failure is unglamorous, mostly documentary, easy to defer in any quarter where the cat season has been quiet, and almost impossible to retrofit on a compressed timeline once a treaty conversation has begun to turn.

The CEO Question, Restated

If our AI drifts and our reinsurer walks, are we still a company?

Most boards do not currently have an answer in the file. The carriers that can produce one (a model change log written for the treaty underwriter, a disclosure register dated through the last renewal, an ORSA scenario that models rescission, a runbook that assumes the rating letter, and a model governance policy that names retraining as a treaty event) are the ones that have done the work. The remaining carriers are operating on a clock that the rating committee, the reinsurer's underwriter, and the next cat season collectively control, and that clock is already running against a renewal date that sits on the calendar in legible ink. The CEO who puts the question on the next board agenda buys the company the runway to answer it on the carrier's own timeline rather than on someone else's.

That is the synthesis. One question, five-step chain, three structural facts, five-item agenda. We think it is the most useful thing a regional or specialty P&C board can put on its 2026 calendar before the next cat season opens.

Related

Footnotes

  1. "FedNat's current catastrophe reinsurance program expires on June 30, and FedNat Insurance and Monarch both indicated that without the reorganization, they were unable to secure adequate reinsurance." — Insurance Journal: FedNat to Cancel 56,000 Policies Under Restructuring Plan; Monarch Finds Investor (May 16, 2022) ↩

  2. "On September 27, 2022, FedNat Insurance Company ('FedNat') was ordered into receivership for purposes of liquidation by the Second Judicial Circuit Court in Leon County, Florida. The Florida Department of Financial Services is the court appointed Receiver of FedNat." — Florida Department of Financial Services: FedNat Insurance Company ↩

  3. "The company was formed in 1906 by a group of farmers in Merced County who found it difficult to obtain fire insurance because the area had no firefighting abilities, according to the company's website… Still, the company carried a solid A-minus rating from insurance consultant A.M. Best before the fire." — United Policyholders: Insurer goes bust from Camp Fire with millions in claims unpaid ↩

  4. "On November 30, 2018, California Insurance Commissioner Dave Jones filed a petition to place Merced Property and Casualty Company ('Merced') into liquidation. On December 3, 2018, the Superior Court of the State of California, County of Merced [entered] an Order Appointing Liquidator." — California Department of Insurance: Regulator takes control of small failing insurer (Press Release release141-18) ↩

  5. "United Property & Casualty had expected gross losses of $660 million from Hurricane Ian, but the actual losses were $864 million. After factoring in reinsurance, the higher-than-expected net loss was $145 million as of Dec. 31, the Securities and Exchange Commission filing said." — Insurance Journal: Orderly Runoff Didn't Work; Florida's United P&C Now Insolvent, Headed for Liquidation (February 20, 2023) ↩

  6. "UPC has declared insolvency in Florida after its estimated losses from Hurricane Ian increased to $1.5 billion, over $500 million higher than the insurer previously projected." — KATC: UPC is insolvent: What homeowners' policy holders need to know (February 2023) ↩

  7. "On February 6, 2023, UPCIC submitted pro forma financial statements to OIR indicating its surplus was ($217,603,217) for the period ending December 31, 2022. This surplus was the result of increased loss reserves and indicated the company was insolvent." — Florida Department of Financial Services: Initial Insolvency Report — United Property and Casualty Insurance Company (2023) ↩

  8. "State Farm General Insurance Company®, State Farm's provider of homeowners insurance in California, will cease accepting new applications including all business and personal lines property and casualty insurance, effective May 27, 2023… This decision was made due to historic increases in construction costs outpacing inflation, rapidly growing catastrophe exposure, and a challenging reinsurance market." — State Farm Newsroom: State Farm General Insurance Company®: California New Business Update (May 26, 2023) ↩

  9. "State Farm General Insurance Co. said it will non-renew 30,000 California homeowners, rental dwelling, and other property insurance policies… These actions are California-specific and will occur on a rolling basis over the next year, beginning on July 3, 2024." — Insurance Journal: State Farm Nonrenewing 30K California Homeowners, Renters (March 20, 2024) ↩

  10. "aiSure™ is a suite of comprehensive coverage for AI systems designed to address a wide area of AI-related risks for AI providers and corporate adopters caused by AI performance errors… aiSure is designed to reflect the probabilistic nature of AI, where even well-constructed models can produce incorrect outputs." — Munich Re: aiSure™ — More AI Opportunity. Less AI Risk ↩

  11. "Swiss Re and RIQ have signed a Memorandum of Understanding (MoU) to develop and scale innovative capacity solutions, risk origination opportunities and AI-enabled capabilities across the UAE… The MoU was signed at Abu Dhabi Finance Week 2025 in the presence of His Excellency Dr Sultan Ahmed Al Jaber, UAE Minister of Industry and Advanced Technology, and Chairman of RIQ, by Mark Wilson, Chief Executive Officer at RIQ, and Andreas Berger, Group Chief Executive Officer at Swiss Re." — Swiss Re: Swiss Re and RIQ partner to advance risk transfer powered by data and AI (December 10, 2025) ↩

  12. "AM Best's 'US Property/Casualty Downgrades Outpace Upgrades in 2023' report noted there were 55 downgrades last year — a higher total compared to upgrades (35) in 2023, as well as to the number of downgrades (30) in 2022." — Insurance Journal: Twice as Many Personal Lines Insurers Downgraded by AM Best in 2023 (April 2024) ↩

  13. "Demotech sent a letter to 17 Floridian insurers on Tuesday explaining that they are likely to have their A ratings downgraded on July 26th… The letter warned 17 carriers in the state that they would be downgraded to S 'substantial' or M 'moderate' as they do not meet the requirements to sustain an A 'unsurpassed' or 'exceptional' financial strength rating." — Artemis.bm: Demotech warns 17 Florida carriers of downgrades, state leadership responds and Insurance Journal: Florida Regulators, FAIA Slam Demotech for Reported Plans to Downgrade 17 Carriers (July 21, 2022) ↩

  14. "The Second Circuit also vacated and remanded on the cross-appeal as the plaintiff did not demonstrate its entitlement to a judgment that the defendant was bound to indemnify the plaintiff according to the plaintiff's settlement with the insured." — The Daily Record (NY): Second Circuit – Reinsurance: Utica Mutual Ins. Co. v. Clearwater Ins. Co. (October 16, 2018) ↩

  15. "Net loss ratio, the ratio of the net losses and loss adjustment expenses to the net earned premium, was 273%, which was 23 points higher compared to Q1 2022." — Coverager: Hippo's Q1 2023 results (May 2023) ↩

  16. "Last year, the hottest on record, secondary perils accounted for 86% of global insurance losses, according to insurance broker Aon Plc." — Insurance Journal: Catastrophe Bonds Use Models Underestimating Climate Risks, Investors Say (May 13, 2024) ↩

Related Posts

Join our newsletter for AI Insights