The governance wedge · Article 5 of 7
Your liability policy may already exclude AI
Insurers have started writing generative AI out of standard liability policies, one renewal at a time. The safety net many firms assume is there may already be gone.
1 · The quiet end of silent coverage
For the first few years of the generative AI era, most businesses enjoyed a kind of accidental insurance. Their commercial general liability and errors-and-omissions policies were written before chatbots existed, so AI-related mishaps were often covered by default. Not because an underwriter priced that risk. Because nothing in the form excluded it. The industry has a name for this, and the name is silent coverage: coverage that exists because the policy has not yet learned to say no.
Silent coverage has one reliable property. It ends. Insurers do not leave unpriced risk on their books once they can see it, and they can see this one now.
The correction does not arrive as a headline. It arrives as an endorsement page in your renewal packet, which is the part of the packet nobody reads.
2 · The forms have numbers now
This stopped being speculative in 2026. ISO, the Verisk unit whose standard forms underpin most U.S. commercial policies, has introduced endorsements carving generative AI out of general liability coverage, and trade press from August 2026 describes carriers adopting them, along with their own versions, at increasing speed. Three form numbers are worth recognizing if they turn up in your renewal.
| Endorsement | What it excludes, as described in trade coverage |
|---|---|
| CG 40 47 | Bodily injury, property damage, and personal and advertising injury arising out of generative AI, for the commercial general liability coverage part |
| CG 40 48 | Personal and advertising injury arising out of generative AI, for the commercial general liability coverage part |
| CG 35 08 | Bodily injury and property damage arising out of generative AI, for the products and completed operations coverage part |
The reporting around these forms is blunt. A coverage attorney quoted in Insurance Journal in August 2026 called it “a major shift in the insurance industry’s treatment of AI-related risks,” with insurers “moving very quickly to limit this exposure.” The same article notes at least one carrier introducing an absolute AI exclusion as early as 2025, and cites litigation-tracking figures, reported there rather than verified by us, showing AI-related lawsuits growing several-fold since 2021. Direction is the takeaway rather than precision. The market is repricing this risk right now, and exclusions are how insurance markets reprice.
Two cautions before anyone forwards this to a partner meeting. An ISO endorsement existing does not mean your policy carries it, because endorsements attach policy by policy, renewal by renewal, and plenty of policies today carry none. And we are technology consultants rather than brokers or coverage counsel, so nothing here is insurance advice. What we can tell you is what the pattern means operationally for a firm adopting AI, and what to ask the people whose advice it is.
Meanwhile, consider the position a typical professional firm now occupies without knowing it. On one side, AI adoption is accelerating inside the building, sanctioned or otherwise. On the other, the insurance that would have quietly absorbed an AI-related claim is being withdrawn form by form. The firm walks toward the risk while the net is rolled up, and both movements are gradual enough that nobody schedules a meeting about either.
For reputation-first firms the exposure concentrates in the deliverable. A fabricated citation in a filed brief. A hallucinated figure in a published reference work. An AI-drafted analysis that misstates a client’s position. These are professional-liability events, errors-and-omissions territory, and the trade press reports insurers reviewing that class for AI language alongside directors-and-officers and fiduciary lines. If such a claim arrived next quarter, the question would not be whether your firm uses AI carefully in general. It would be what your policy says, and then, immediately, what you can document about how the work was produced.
3 · Six questions for your broker, and why the fifth is the one
The operational move costs one conversation. Put these in writing, before your next renewal cycle rather than during it.
- Do any of our current policies carry AI-related exclusion endorsements, and are any proposed for the coming renewal, on any line: general liability, errors and omissions, directors and officers, cyber, media liability?
- How is “generative artificial intelligence” defined in any such endorsement, and would it reach the AI features embedded in ordinary software we already use, as opposed to standalone chatbots?
- Where are we relying on silent coverage today for AI-related scenarios, and which of those would the pending ISO forms remove?
- Does our errors-and-omissions policy speak to AI-assisted professional services at all, affirmatively or by exclusion?
- Would documented AI governance change our position, in underwriting questions, pricing, or insurability, and what documentation would carriers want to see?
- Are affirmative AI coverages or riders available in our lines and size class, and what do they cost?
A broker who answers these crisply is earning their commission. A broker who waves them off has told you something too.
Question five is where this article joins the rest of the series. When underwriters start asking how a firm governs AI, and the trade coverage suggests they will, the firms that answer well will be the ones holding artifacts: a written risk-tiered policy with acknowledgments, a current tool inventory with account tiers, named review accountability, contracts checked against usage. Which is, item for item, the checklist this series has been walking.
Those same documents do double duty at claim time. Whatever your policy covers, the defense of a professional-liability claim involving AI will turn on whether the firm behaved like a firm with a system, and the distance between asserting that and producing it is the distance between an argument and an exhibit. Insurance is the one audience in this series you cannot charm in the moment. It reads paper, priced in advance.
The checklist domain this article extends is the contracts one, and its last box is the bridge:
☐ We’d be comfortable putting our AI use on an account in writing.
Tick that one truthfully and the broker conversation gets easy, and the renewal becomes what renewals should be, which is boring.
Where this fits
The next article turns from paper to people: human oversight, what “reviewed by a qualified person” has to mean now that courts are sanctioning professionals whose review was a glance, and how to build gates that survive a busy week.