Who Pays When the Robot Is Wrong?

Kim opens this episode holding a letter that has not been written yet. It is a claim denial addressed to a small business owner whose AI tool made a promise the business could not keep, citing the generative AI exclusion that quietly landed inside most business insurance policies in the country this January. Nobody has actually received that letter. But the exclusion is real, and almost nobody who turned on an AI receptionist, a quoting tool, or a two-in-the-morning chat window this year has read it. Kim and Hal spend the episode answering the question underneath the letter. Not whether to use AI, they settled that a long time ago, but who pays when it is wrong. Hal takes the buyer's chair. An under-automated business isn't safe, it's just cheap, and an uninsured AI incident is a hole in the balance sheet that shows up the day you try to sell, refinance, or sleep at night. Kim explains why carriers wrote the exclusion in the first place: they can't yet price a mistakethat could repeat across a million businesses on the same afternoon. And she walks through why "the software did it" stopped being a defense the moment a tribunal ruled, in the Air Canada chatbot case, that a company owns what its bot says on its own website. The fix is smaller than it sounds. Using the reversibility rule from a few episodes back, Kim and Hal draw the line at exactly three doors an AI agent should never open alone: moving money, making a binding promise, and committing the business legally. Everything else, let it run at full speed. Kim also credits risk advisor Gregory Marholin, Executive Advisor for Strategy and Growth at AI GuardWorks, whose piece for the Big I's Virtual University makes the same case from inside the insurance industry: a vendor list answers whoyou're paying, an AI inventory has to answer where AI is making or influencing a judgment. And here's the honest part. A specialist AI liability insurance market is forming this year, but it's new, it isn't standardized, and it isn't cheap yet. The move isn't to freeze. It's to know your exclusion, list your exposure, and gate the three doors that matter. "That's how you build a technology-enabledsuperpower," as Kim puts it: not that the tool is smart, but that you're covered, gated, and deployed anyway. **In this episode:** ●      The fictional denial letter, and the real endorsement behind it ●      Why carriers excluded generative AI from standard business policies starting in January 2026 ●      The Air Canada chatbot ruling, and why "that's the vendor's problem" stopped working as a defense ●      Why freezing your AI rollout doesn't protect your business, it marks it down ●      The reversibility rule: the three doors that always need a human hand ●     Gregory Marholin's "vendor list versus AI inventory" framing, from his piece for the Big I Virtual University ●      The specialist AI liability insurance market taking shape in 2026, and why it doesn't replace knowing your existing exclusion ●      The three-step agent liability audit you can run before lunch **Ready to take action:** Website and IMPACT: smallbusinessbigai.com The IMPACT Framework: smallbusinessbigai.com/impact Kim Lewis Howard on LinkedIn (send your story here) Hal Howard on LinkedIn Gregory Marholin's article, Big I Virtual University Insurance assessment, commercial or personal lines: asklewishoward.com