As of January 1, 2026, insurers can exclude losses from generative AI out of design-professional errors-and-omissions policies. If your firm uses AI to draft SF330 content and that content overstates a qualification, the resulting claim may no longer be covered. That turns an AI drafting shortcut from a proposal-quality question into a professional-liability one. This is a principal's problem, not a coordinator's.
What Changed on January 1
Verisk released two standardized AI exclusion endorsements, CG 40 47 and CG 40 48, giving carriers ready-made language to carve generative-AI losses out of coverage. Several carriers moved quickly. Philadelphia Insurance and Hamilton Select have been the most aggressive, excluding AI-related claims outright, and others including AIG, Great American, and W.R. Berkley have filed for similar exclusions. This isn't a future scenario being debated. It's showing up in 2026 renewal cycles right now.
The exclusions generally come with a carve-back. A firm that can document an internal AI governance program, meaning tool approval, defined verification steps, client disclosure where appropriate, and licensed-professional sign-off, can often preserve coverage. The coverage doesn't disappear because AI was used. It disappears when AI was used without a documented process behind it.
None of this is legal advice, and the specifics turn on your actual policy language. The point a firm owner should take from it is narrower: the insurance treatment of AI-assisted work has changed, and a conversation with your broker before the next renewal is overdue.
Why an SF330 Is the Exposure
Most discussion of E&O and AI focuses on design deliverables, the calculations and drawings. The proposal is the part firms overlook, and it may be the more immediate exposure.
Here's the mechanism. A qualifications package is a set of representations. When you submit an SF330, you're representing that the named people hold the credentials listed, that they served in the roles described, and that the projects happened as written. AI drafting introduces a specific failure mode into exactly those representations. As covered in the pre-submission verification pass, the danger isn't fabricated nonsense. It's confident overstatement: a role promoted past what the person actually did, a certification added because the role usually carries one, a project scope inflated.
When that overstatement goes out under signature, it stops being a proposal-quality issue and becomes a misrepresentation of professional qualifications. If a dispute later turns on it, the firm is defending a statement it made about its own people and work, generated by a tool, possibly under an exclusion that now applies. That's the scenario the new endorsements were written for. The standard-of-care question for AI-drafted submittals was already live; the insurance change raises the cost of getting it wrong.
What the Carve-Back Actually Requires
The good news is that the documented-governance carve-back isn't exotic. It's the verification discipline a careful firm should already run, written down. Four elements show up across carrier requirements:
-
Tool approval. The firm decides which AI tools are permitted for proposal work rather than letting each person use whatever they prefer. An undocumented patchwork of tools is hard to govern and harder to insure.
-
Defined verification steps. Every AI-drafted claim about a person or project is traced back to a source record before submission. The role, the years, the certifications, and the project facts are confirmed against what the firm can actually show.
-
Client disclosure where appropriate. Some agencies and clients now ask how AI was used. Having a consistent answer is part of a defensible process.
-
Licensed-professional sign-off. A licensed professional reviews and stands behind the qualifications content before it goes out, the same accountability that applies to any other professional representation the firm makes.
A firm that does these four things and keeps a record of doing them is in a very different position at renewal, and in a dispute, than one that ran AI through a generic chat box and signed whatever it produced.
The Practical Move
The verification steps in the carve-back are easier to run when the firm's qualifications already exist as structured, auditable data rather than scattered across Word files and individual memory. If every claim in a package traces to a source record of staff and project facts, the verification step is a confirmation against the record rather than a hunt. That's the governance carve-back in practice: AI drafts from data the firm can show, and the source is auditable. It's also why AI grounded in a firm's own records behaves differently from a general-purpose tool that fills gaps with what's statistically likely and confidently overstates.
The takeaway isn't to stop using AI on proposals. It's that after January 1, using it without a documented verification process carries a liability the firm may now be carrying uninsured. Build the process, write it down, and talk to your broker before the next renewal.
Frequently Asked Questions
Can E&O insurance exclude AI-related claims now?
Yes. As of January 1, 2026, Verisk's CG 40 47 and CG 40 48 endorsements give carriers standardized language to exclude generative-AI losses from design-professional E&O policies. Philadelphia Insurance and Hamilton Select have excluded AI claims outright, and other carriers have filed for similar exclusions. Many policies offer a carve-back for firms with a documented AI governance program.
Is AI-drafted proposal content a professional liability risk?
It can be. A qualifications package is a set of representations about your people and projects. If AI overstates a role, certification, or project scope and that content goes out under signature, it becomes a misrepresentation of professional qualifications rather than just a weak proposal. Under the new exclusions, a resulting claim may not be covered.
What does an AI governance program require for E&O coverage?
Carrier carve-backs generally require four things: approving which AI tools are used, defining verification steps that trace every AI-drafted claim to a source record, disclosing AI use to clients where appropriate, and having a licensed professional sign off on the content. Documenting that you do these is what preserves coverage.
Does this apply only to SF330 proposals?
No. The exposure applies to any qualifications content where AI overstates a credential or project, including SOQs, state DOT prequalification packages, and RFQ responses. SF330 is a clear example because its representations are formal and signed, but the same misrepresentation risk exists in any AI-drafted package the firm submits.
Should I stop using AI on proposals because of the exclusions?
No. The exclusions target undocumented use, not AI itself. Firms that approve their tools, verify AI-drafted claims against source records, and have a licensed professional sign off can usually preserve coverage through the carve-back. The risk is using AI without a process. Build and document the process, and confirm your specific policy language with your broker.