In late 2025, several major professional-liability carriers — including names that write E&O coverage for law firms, title companies, financial advisors, and engineering consultancies — quietly amended their policy language. Specifically, they carved out coverage for losses arising from the use of “generative artificial intelligence.” It did not make the front page of the trade press. It barely made the back page.
It should have been front-page news. What those carriers did, in a single fall, was identify the largest professional-liability gap of the decade. And almost nobody in the technology industry noticed.
What the Carve-Out Actually Means
The standard professional-liability policy protects a licensed professional against claims arising from an error in their work. A lawyer drafts an opinion letter that misses a controlling case; a title officer issues a commitment that overlooks an unreleased deed of trust; an engineer signs a structural calculation that depends on a flawed assumption. Claim arises, carrier pays, professional keeps practicing.
The carve-out says: if any part of that work product was generated, drafted, summarized, or assisted by a generative AI system, the carrier may decline the claim entirely.
That sounds narrow. It is not narrow. Generative-AI assistance is now ubiquitous in the professional workflow — drafting correspondence, summarizing depositions, surfacing precedent, condensing technical documentation. Every licensed professional using a 2025-era AI tool to assist their work is now operating at least partially outside their coverage perimeter. They may not have been told that. Most have not been told that.
What the Carriers Saw That Builders Missed
The artificial-intelligence industry continues to argue about hallucination rates as if hallucination is the problem. Lower hallucination, the argument goes, and the technology becomes safe. The carriers do not share this view.
If a model produces an answer that is wrong only one time in a thousand, and the model is producing a thousand answers a day, then somewhere in the workflow there is a wrong answer every day — and no one knows which one.
That is not a hallucination problem. That is a verification problem. It is the same problem aviation solved with checklists, accounting solved with audit, and medicine solved with peer review. You cannot eliminate error from a high-volume judgment process. You can, however, build a verification layer that catches error before it becomes consequence.
The carriers — who price risk for a living — looked at the existing AI stack, did not find a verification layer, and concluded that the responsible underwriting decision was to carve the entire category out of coverage until one appears.
Why This Is a Sparked Observation
The professional-liability gap is not a niche compliance question. It is the canary signal that the entire AI industry has built the wrong half of the stack. We have spent four years optimizing models and almost no time at all building the verification, provenance, and accountability infrastructure that would let those models be trusted in serious work.
That gap is the work in front of us. Verification is the bridge between capability and coverage — between what an AI can do and what a licensed professional can sign their name to. The first builders to close that gap will define the next decade of how AI gets used in everything that actually matters.
The carriers have already named the problem in the language they use for naming problems: a policy exclusion. The rest of the industry is one underwriting cycle away from being forced to listen.
Shawn Paul Cosner
Sparked Technology Solutions, Inc.