Explore the Advocate app here: https://advocate.app

AI-related damages are quietly being carved out of commercial general liability policies. Three of the largest carriers in the country, Chubb, Berkshire Hathaway, and Travelers, just got the green light from state regulators to start excluding AI from standard CGL coverage. ISO released two new AI exclusion endorsements that went live January 1st. More than 80% of these requests are getting approved. And it's barely been covered in the news.

The number on your declaration page is not the coverage. The exclusions, the endorsements, and the new language being filed underneath you are the coverage.

In this episode of the Advocate Insurance Desk, Katie and Grace break down what just happened in the commercial liability market and why it's the silent cyber playbook running a second time. Then they bring on Marek, Advocate's Head of Infrastructure and Security, to talk through what AI risk actually looks like from the inside: compounded supply chain attacks, the new Anthropic model that finds and exploits vulnerabilities on its own, why he denied a request to give an AI assistant access to Outlook, and the small annoying things every operator should be doing Monday morning.

We cover:

How Chubb, Berkshire Hathaway, and Travelers got regulator approval to exclude AI-related damages from standard CGL policies in less than four months

Why the speed of this carve-out, regulator approved and carrier deployed in a fraction of the usual time, signals how worried the market actually is

The ISO endorsements that went live January 1st and what they actually exclude: defamation from AI output, IP infringement from AI generated content, and physical damage traced back to AI error

A real supply chain attack on an NPM library that exfiltrated developer secrets without any user action, and why this is the failure mode carriers are scared of

Anthropic's new Mythos model, only released to about ten of the biggest tech companies, and what it signals about where AI risk is heading

Why Marek denied a request to give an AI assistant access to Outlook, and how he thinks about department-level AI governance for sensitive data

The story of an AI tool that destroyed a production database including the backups, and what it tells you about agentic access

The buyer-broker gap: most clients can't answer where AI is being used in their own operations, and most brokers aren't tracking how carrier policy language is shifting underneath them

Why broader CGL coverage isn't coming back, and what new standalone AI products entering the market actually need to look like to fill the gap

The questions every operator should be asking their broker before their next renewal

If you own, operate, broker, or underwrite anything that uses AI in any part of its tech stack, and that's almost everyone now, this episode gives you the frame for what just changed and what to ask before your next renewal.

0:00 Introduction

0:43 Three Carriers Just Walked Away From AI Risk

1:37 The ISO Endorsements That Went Live January 1st

2:22 Why This Moved So Fast

2:42 What a CGL Policy Actually Covers

3:04 If You Use AI, You're Exposed

4:37 Bringing on Marek, Head of Infrastructure and Security

6:02 The NPM Supply Chain Attack

7:12 Compounded Aggregated AI Risk

9:15 Anthropic's Mythos and What's Coming

11:02 Open Source vs Closed Source in the AI Era

11:49 What Financial Institutions Worry About in Due Diligence

14:05 Why Marek Denied the AI Outlook Request

15:29 The Annoying Things Operators Should Do Monday Morning

16:28 When AI Destroys a Production Database

17:20 Wrapping with Marek

19:17 The Buyer Is Stuck and the Broker Gap

22:16 Why Broader CGL Coverage Isn't Coming Back

22:38 The Path Forward: Standalone AI Products

24:28 Four Questions to Ask Your Broker Right Now

24:50 Visibility First, Coverage Second

25:40 Outro

#AI #Insurance #CommercialInsurance #RiskManagement