Show Notes
Everyone's pricing AI as a headline risk. Jeremy Epstein thinks that's the wrong question entirely — the risk isn't that a company uses AI, it's what authority they've quietly handed it to act without anyone watching. He left underwriting for venture, watched every pitch deck turn into "AI-native law firm," and built Mayflower Specialty to insure the gap nobody else is pricing.
WHAT YOU'LL LEARN: Why "AI" as a noun tells you nothing about the risk — and what to ask instead How a five-person team is underwriting exposure for organisations backed by AA-rated capital The line between AI drafting a suggestion and AI shipping it unsupervised — and why that line is the whole policy Why standards like NIST and ISO 42001 are becoming the difference between insurable and uninsurable What building "from zero" in the AI era actually looks like versus retrofitting a legacy insurer
Everyone's pricing AI as a headline risk. Jeremy Epstein thinks that's the wrong question entirely — the risk isn't that a company uses AI, it's what authority they've quietly handed it to act without anyone watching. He left underwriting for venture, watched every pitch deck turn into "AI-native law firm," and built Mayflower Specialty to insure the gap nobody else is pricing.
WHAT YOU'LL LEARN: Why "AI" as a noun tells you nothing about the risk — and what to ask instead How a five-person team is underwriting exposure for organisations backed by AA-rated capital The line between AI drafting a suggestion and AI shipping it unsupervised — and why that line is the whole policy Why standards like NIST and ISO 42001 are becoming the difference between insurable and uninsurable What building "from zero" in the AI era actually looks like versus retrofitting a legacy insurer
EPISODE LINKS: Jeremy's LinkedIn: https://www.linkedin.com/in/jaepstein00002/ Mayflower Specialty: https://mayflowerspecialty.com/
OUTLINE & TIMESTAMPS: 00:00 Meet Jeremy & the Idea Behind AI Insurance 00:52 Building an AI-Native Insurance Business 02:08 Running an MGA With Just Five People 03:02 When Does AI Stop Replacing Headcount? 03:52 The Real Risk: What You Let AI Do 05:05 Underwriting AI Deployment 06:05 AI Isn’t Just Software Anymore 07:05 What Traditional Insurers Can Learn From AI-Native Firms 07:42 Final Thoughts
More from The Reinsurance Podcast
Dame Susan Langley, the first Lady Mayor of the City of London in 800 years and only the third woman to hold the office, has spent a year flying round the world telling people what London’s insurance and reinsurance market actually does. Recorded at Mansion House, this is her case for why the quietest industry in finance needs to find its voice.
Sub-$50bn of first-half cat losses, no hurricanes last year, none so far this year, and rates heading south. Verisk's Rob Newbold boards the TRP yacht at Monte Carlo RVS to explain why reinsurance should still be braced for $171bn of insured catastrophe loss in an average year. Then he gets into what AI is actually doing to cat modelling, from globally correlated models to agents that talk to each other so humans can get on with deciding.
New capital is still queuing up for Lloyd's, but Argenta's Kate Tongue and Robert Flach reckon the smart money is taking its time. On board the Supercede yacht at Monte Carlo RVS, they explain three things: why investors tend to show up a year too late, why a diversified Lloyd's portfolio can rival ILS returns with less risk, and why capital cares less about a syndicate's AI than about who's parameterising it.
Property is softening at a sprint. Casualty is walking, and keeping one eye on US social inflation. On the TRP yacht at Monte Carlo RVS 2026, Lockton Re's Carla Moffett and Sam Sweeney explain why casualty reinsurance won't follow the rest of the market down, and why the brokers worth hiring now do much more than place the treaty.
Rachel Turk, Chief of Performance and Strategy at Lloyd's, explains why the flood of third-party capital is good news only while there's fresh risk for it to chase, and why institutional money might one day walk for reasons that have nothing to do with the insurance cycle. Plus: data centres that want $20bn of cover but not the bill, and why agentic AI underwriting only works if it explains itself in plain English.