In Depth
A policy is built in layers. The insuring agreement grants broad coverage, the definitions set its boundaries, and the exclusions subtract the risks the insurer never intended to carry. The real scope of a policy is not what the cover page promises. It's what survives after the exclusions are applied, and that's where most coverage disputes live.
Exclusions exist for sound reasons. Some risks are better handled by another policy (a CGL form excludes professional services because that's what E&O is for). Some are uninsurable, or priced into a separate product. And some get added the moment an insurer realizes a new risk is showing up in claims it never underwrote for. AI is the textbook case of the third kind. As generative-AI losses started surfacing, carriers moved quickly to write AI-specific language into Tech E&O, Cyber, and CGL forms, not because AI is uninsurable but because their existing forms hadn't priced for it.
The danger for a vendor is that an exclusion stays silent until a claim tests it. You pay premiums, you hold a certificate, you assume you're covered. Then a hallucination-driven claim arrives, the adjuster reads the AI carve-out added at the last renewal, and the denial letter explains that this particular failure was excluded all along. Nothing about the policy looked different. The coverage was already gone.
What It Looks Like
A vendor renews its Tech E&O policy. The renewal packet runs to forty pages. The broker highlights the premium and the limits, and the vendor signs. Buried in the endorsement section is a new clause excluding "any claim arising out of or related to the use of artificial intelligence or machine-learning systems." Six months later the vendor's agent gives a customer wrong guidance, the customer sues, and the vendor files the claim. The carrier points to the endorsement. The policy is in force, premiums are paid, and the single risk the vendor most needed covered is the one the renewal quietly removed.
Why It Matters For AI Vendors
Exclusions are how the coverage gap gets created in the first place. An AI vendor can carry Tech E&O, Cyber, and CGL, pass its customer's insurance requirements, and still be uncovered for hallucination, bias, prompt injection, and data disclosure, because each of those forms now carries an AI carve-out written specifically to keep those claims out.
The practical lesson is to read the exclusions before the limits. A high limit on a policy that excludes how your agent actually fails is worth less than a smaller limit on a policy that covers it. The exclusion page is where you find out which one you're holding.