The Record · Part 1 of 5
The Model Will Say Yes
Who owns AI output, and is your insurance still silent on it?
A project engineer opens a public chatbot and asks it to check a crane pick plan against the load chart. The answer comes back in seconds, clear and confident. (This scenario is illustrative.)
The tool never asked whether the engineer should rely on that answer. These tools are built to be helpful. If a request isn’t obviously illegal, the model will usually try to answer it.
That’s the idea behind this five-part series. AI tools say yes to almost anything. The contractor carries the risk. And in 2026, the insurance market started writing it into policies.
The vendor’s fine print
Read an AI tool’s terms and the message is clear: checking the output is your job.
Take OpenAI, the maker of ChatGPT. Its consumer terms say, “You must evaluate Output for accuracy and appropriateness for your use case.” Its business agreement makes the customer “solely responsible for all use of the Outputs and for evaluating the accuracy” of them.
What changes between a personal and a business plan is money and data:
- Liability cap (the most the vendor will owe you). On a personal account, OpenAI’s liability tops out at whichever is greater: 12 months of fees or $100. On a business plan, it’s 12 months of fees, with some exceptions.
- Training. OpenAI can use personal-account content to train its models unless the user opts out. It doesn’t train on business content unless the customer agrees.
OpenAI’s usage policies also assume a professional is in the loop. Two of their rules:
- No “tailored advice that requires a license” without “appropriate involvement by a licensed professional.”
- No automating high-stakes decisions without human review in areas that include “critical infrastructure” and “product safety components.”
Those rules are between you and the vendor. They don’t set a standard of care, the legal yardstick for what a careful professional would do. And the tool doesn’t check who’s supervising.
Construction is moving fast
This isn’t a future problem. In the AGC and Sage 2026 outlook, 61% of firms said they use AI or plan to invest more in it, up from 44% in the previous outlook. Nearly a quarter use it for estimating. One in five use it for design and preconstruction.
Insurance is no longer silent on AI
For years, most policies said nothing about AI. The industry calls that “silent AI.” It’s ending, and each type of policy is moving its own way.
General liability: new optional exclusions. CGL (commercial general liability) is your main liability policy. Verisk writes the standard forms many carriers use, known as ISO forms.
It released three optional AI exclusions, effective January 2026. Each is an endorsement, an add-on form that changes what a policy covers:
- CG 40 47 removes coverage for bodily injury, property damage, and personal and advertising injury arising out of generative AI.
- CG 40 48 removes only personal and advertising injury (claims such as libel or invasion of privacy).
- CG 35 08 removes bodily injury and property damage arising out of generative AI under the separate products and completed operations coverage part.
Optional is the key word: each carrier decides whether to use them. More than 60 property and casualty insurers filed AI exclusions in 2026, according to CSIS.
Filing a form with state regulators doesn’t put it on your policy. But Gallagher told ENR the ISO endorsement is “beginning to appear on” CGL policies.
Back to that crane lift. Say it fails and someone is hurt. If your CGL includes CG 40 47, the carrier may argue the injury arose out of generative AI.
That’s an argument, not a certainty. It’s still one you’d rather not have.
Management and professional lines: broader exclusions are already here. W.R. Berkley has an “absolute” AI exclusion. Reports place it on D&O and fiduciary forms, plus E&O or EPL (sources differ). Those policies cover your directors and officers (D&O), professional errors and omissions (E&O), employment practices (EPL) and employee benefit plans (fiduciary).
Hamilton has a generative AI exclusion in professional liability. Berkley’s form, though, isn’t being used on design or construction policies, ENR reports.
Design professional E&O: mostly steady, for now. In October, ENR found no AI exclusions yet on policies for design firms and contractors. AXA XL said its architect and engineer forms are “agnostic as to how the matter occurs.” In other words, the forms don’t care how a claim came about.
But one wholesale broker, PDI, reports that at least one major carrier has added an absolute AI exclusion to E&O.
Cyber: moving the other way. Here, carriers are clarifying coverage instead. Beazley’s US cyber wording now spells out “AI-related risks already covered.”
CFC added affirmative AI cover, meaning language that expressly includes AI, to its cyber wording for financial institutions. Lockton says it is “not aware of AI being excluded on any cyber or technology insurance products.”
At least one major broker is pushing back on the broad versions. On Oct. 15, Marsh CEO John Doyle said a broad AI exclusion “could create real confusion and uncertainty.”
What this means for you
Put the two halves together. The vendor makes checking the answer your job. Whether your insurance backs you up can depend on which forms are attached.
So your exposure comes down to three things:
- Your declarations page, the summary at the front of your policy
- Your forms schedule, the list of every form and endorsement on it
- The AI accounts your people actually use
A form filed in your state means nothing until it’s attached to your policy. And a business AI plan protects your data, but it doesn’t make the vendor responsible for a wrong answer.
Do this month
- Ask your broker to list every endorsement that mentions “artificial intelligence,” “generative,” “machine learning” or “algorithm.” Include GL, products and completed operations, umbrella, professional, cyber and D&O.
- Find out which AI tools your people use, and on which accounts. For each one, note the plan (personal or business), the training setting and the liability cap.
Ask your broker
Put this in writing: “Is CG 40 47, CG 40 48 or CG 35 08, or a carrier’s own version, attached or proposed at renewal? If so, how does it define generative AI?”
Next week
Part 2: shadow AI, the tools your people use without approval. Nothing gets hacked. The data just leaves. And we found no carrier that has said in public whether that’s covered.
Part 2 publishes Tuesday, October 27, 2026, at 7:45 a.m. Mountain. Get it by email · Follow on LinkedIn · RSS feed
Free resource: Our AI in Construction guide includes ten questions to ask any AI vendor before your teams use it. It’s free to read.
Get The Record by email
Each part in your inbox the morning it publishes.
Then the occasional new guide or template for contractors. Unsubscribe in one click.
We confirm your address by email first. Privacy Policy
General information, not legal or insurance advice. Talk to your broker and counsel.
Sources
- OpenAI Terms of Use, effective Jan 1, 2026
- OpenAI Services Agreement, effective Jan 1, 2026
- OpenAI Usage Policies, effective Oct 29, 2025
- AGC and Sage 2026 Construction Hiring and Business Outlook, Jan 8, 2026
- Big I: Verisk generative AI GL exclusions, Oct 21, 2025
- CSIS: The Insurance Industry’s Retreat from AI, Sept 4, 2026
- ENR: AI exclusions won’t soon be popping up in designer liability policies, Oct 9, 2026
- Zelle: The Growing Trend of AI-Related Exclusions, Oct 31, 2025
- Pillsbury: AI Exclusions in Insurance Policies, Apr 13, 2026
- PDI: AI Tools and Your E&O Policy, Apr 22, 2026
- Beazley: affirmative AI cyber cover, Sept 17, 2026
- Insurance Journal: Beazley and CFC AI cover, Sept 17, 2026
- Lockton Market Update: Cyber, July 2026
- Insurance Journal: Marsh CEO on broad AI exclusions, Oct 16, 2026
About The Record
The Record is written by Jared Smith, founder and CEO of SymphonIQ, which builds governed, single-tenant AI for engineering and construction.