Does Agency E&O Cover AI Mistakes? What to Ask Your Carrier

It depends on your policy wording, and the only reliable answer is your carrier's, in writing. A policy that is silent on AI may treat an AI-assisted error like any other professional mistake, but some insurers now attach AI exclusions to liability policies, so ask the carrier tool by tool before relying on cover.

This is not legal or insurance advice, and nothing here interprets any particular policy. It gives you the questions, the wording to look out for and the evidence to collect, so that a conversation with your broker or carrier is specific. A vague "do we use AI?" gets a vague answer; "we use a call-answering assistant that books quote appointments, does the policy treat that as professional services?" gets one you can file.

Follow me on Instagram@sagnikteaches

Why the question has become urgent

Traditional E&O wording was written around human error: the agent who forgot to add a driver, bound the wrong limit or missed a renewal. An AI-drafted coverage summary that tells a client their policy covers something it does not is the same kind of mistake with a different author, and the claim still lands on the agency. What has changed is that insurers are revisiting how they treat AI-related loss. Verisk's ISO published three optional generative AI exclusions with a January 2026 edition date: CG 40 47 (bodily injury, property damage and personal and advertising injury) and CG 40 48 (personal and advertising injury only) for the commercial general liability coverage part, and CG 35 08 for the products and completed operations coverage part. On professional lines, a law firm's insurance-recovery commentary reported in May 2025 that one insurer had introduced a so-called "absolute" AI exclusion intended for its D&O, E&O and fiduciary liability products, and specialist AI liability products have since appeared to fill the gaps. Read the endorsement schedule at every renewal, because the same agency can face an exclusion with one carrier and none with another.

Connect on LinkedInSagnik Bhattacharya

Two cautions. First, the general liability endorsements are a different policy from your E&O; the point is that you need to check each policy, not assume one tells you about another. Second, trade-press summaries differ in emphasis and some of them are old by the time you read them. The policy forms attached to your own schedule are the source of truth.

Subscribe on YouTube@sagnikteaches

Step 1: find out what the agency actually uses

You cannot answer a carrier question accurately if you do not know what staff do. Before writing to anyone, build a one-page register. Walk the agency in four groups: producers, account managers, claims or service staff and the owner. Ask each: which AI tools do you use for work, including free chatbots, browser add-ins, phone apps and anything built into software we already own? What do you put into them? What do you do with the output?

All entries are illustrative. The "human check" column is where exposure usually sits: an AI-written note nobody reads is the same as an unreviewed junior's note, and carriers tend to ask about review, not about the tool brand. For ideas on what is realistic, see seven AI tools for quotes, renewals and calls.

Step 2: the carrier question list, tool by tool

Send the register with questions attached to the specific use, not a general query. The wording below is a starting point; ask your broker to refine it.

Questions that apply to every tool

• Does our E&O policy, or any endorsement on it, contain an exclusion, sublimit or limiting definition referring to artificial intelligence, generative AI, machine learning, automated systems or algorithms? Please list the form numbers.
• Does the policy treat a professional-services error the same way whether a person or an AI-assisted process produced it, provided a licensed person reviewed it?
• Are there conditions, warranties or application statements about how we use technology that could affect a claim?
• Will anything change at the next renewal, and how would we be told?
• If the answer to any of these is "it depends", please describe in writing what it depends on.

Call-answering and chat assistants

• If an assistant tells a caller or website visitor something inaccurate about coverage or a quote before a licensed person is involved, is that within the professional services definition?
• Does the policy treat an assistant's recorded statement as an agency statement?
• Do you require disclosure to callers that they are talking to an automated assistant?

The design of these tools matters as much as the policy. The pattern in an AI chatbot for quote requests, not advice limits what an assistant is allowed to say, which gives you something concrete to show an underwriter.

Drafting tools (emails, summaries, proposals)

• Does the policy cover errors in documents prepared with AI assistance, such as a coverage comparison sent to a client?
• Is there any requirement that a licensed person review client-facing documents?

Tools that hold client data

• Is a data leak caused by an AI tool covered under the E&O, the cyber policy, both or neither?
• Does the cyber policy have any condition about third-party tools or vendors?

The data side connects to your information-security duties, covered in the safeguards checklist for AI tools. Carriers sometimes treat a breach and an error as different claims, so ask both questions.

Wording to look for when the policy comes back

You are not trying to become a coverage lawyer, but a few features of the wording are worth marking up and asking your broker about.

The claims-made row deserves a moment. Under a claims-made policy, what counts is generally the policy in force when the claim is made and notified, subject to the retroactive date. So an exclusion that appears at renewal can matter for work done under last year's policy if the claim arrives after the change. That is exactly the kind of thing to ask the broker, in writing, and not to assume.

An example: the quote-comparison error

Say a small commercial agency uses a general chatbot to draft side-by-side comparisons of two quotes for a restaurant client. The model summarises the policies and states that both include equipment breakdown. One does not. The client suffers a loss, finds the gap and complains. The mistake could be described as an error in a coverage comparison (a classic E&O scenario) or as an AI output failure. How the policy and the carrier treat it will turn on the wording, on whether a licensed person reviewed the comparison, and on what the agency told the carrier at application.

Now see what evidence helps: a written AI policy that says comparisons must be checked against the policy documents by a licensed person; a log showing the account manager did that check, or a note that the check was missed; a tool register showing the chatbot was an approved use. None of that guarantees cover, but it shows control, and a carrier may weigh it in a conversation. It also separates a process failure that you can fix from a systemic one.

A written AI policy as evidence

A one-page policy is worth having whatever the carrier says, and it answers the underwriting questions that appear on applications and renewals. Keep it short enough that staff read it. A skeleton, filled in with illustrative content:

AGENCY AI USE POLICY (v1, reviewed [date])

1. Approved tools: [Tool A for drafting; Tool B for call answering;
   Tool C inside the agency management system]. Anything else needs
   the principal's written approval.
2. Never enter: full policy numbers, payment card details, health
   information, or claim details that identify an individual.
3. Client-facing documents: a licensed person reads and approves
   every AI-assisted coverage summary, comparison or proposal
   before it is sent. The approver's initials go in the file note.
4. Assistants talking to the public: scripted to collect details
   and book calls; they never state what is or is not covered.
5. Errors: any suspected AI-related error is reported to the
   principal the same day and logged.
6. Review: the tool register is checked quarterly and before each
   renewal.

You can use AI to turn rough notes into this document, and turning routines into written procedures with AI shows how. Do not let it invent controls you do not actually operate. A policy the agency does not follow is worse than none, because it becomes a record of what you said you did.

A draft email to the carrier or broker

This is a reasonable first message. Adapt it, and ask the broker to send it if they handle the relationship.

Subject: AI use at [agency] and our E&O policy [policy reference]

Hello [first name],

Ahead of renewal I'd like written confirmation of how our E&O
policy treats errors in work prepared with AI assistance.

Attached is a one-page register of the AI tools we use, what
they are used for and the human review that applies. Could you
confirm:
1. whether any exclusion, sublimit or limiting definition on the
   policy refers to AI or automated systems (form numbers please);
2. whether work prepared with AI assistance and reviewed by a
   licensed person is treated as any other professional service;
3. what we should tell you if we adopt a new AI tool mid-term;
4. whether any of this changes at renewal.

Thank you,
[your name]

File the reply with the policy. If the answer is evasive, that is information too, and a reason to ask a specialist broker to compare alternatives at renewal.

What to do if the answer is bad news

• A broad exclusion with no exceptions: list which uses it touches. Often the realistic response is to restrict those uses (for example, no AI-drafted coverage comparisons) until cover is agreed, or to ask the broker to find a policy without the exclusion.
• A sublimit: ask what it would take to raise it, and whether the exposure is large enough to matter given your client mix.
• "We can't say": ask for a named underwriter contact and put the question in writing again with the register attached.
• Separate AI cover offered: some carriers are starting to offer dedicated AI endorsements or products. Compare the wording and price against the risk and against how you use the tools. Ask what it covers and excludes in as much detail as for your main policy.

Keeping the evidence current

A policy and a register go stale quickly, because staff adopt new features without announcement. Two light habits keep it current. Each quarter, ask staff the same three questions as in Step 1 and update the register; and add a standing item to the monthly meeting: "any new AI feature switched on in a system we use?" Many agency management systems and phone systems roll AI features into updates, so the tool may not be new to you but its behaviour is. Before you add a tool, weigh its first-year cost against the extra register, policy and carrier work it brings.

If you are also relying on an outside consultant to set tools up, the E&O angle on that relationship is in hiring an AI consultant: scope and E&O. And for sector rules on AI use by licensed producers, the AI model bulletin, explained sets out the expectations a carrier may also have in mind.

If you would like help turning this into your own register, policy and carrier question sheet, that is what my AI implementation consultation covers: get in touch. Otherwise, start with the register, because everything else follows from knowing what is actually in use.

A renewal calendar for the AI conversation

Timing is half the battle, because a carrier's answer carries more weight when it arrives before the renewal terms are set. A sensible sequence for an agency with an annual E&O renewal looks like this.

1. Ten to twelve weeks out: update the tool register, check for new AI features in the systems you own, and read last year's endorsement schedule for any AI-related forms.
2. Eight weeks out: send the question email to the broker with the register and the draft AI policy attached.
3. Six weeks out: when the renewal application arrives, compare every technology question on it against the register. Answer with the facts in the register, not from memory.
4. Four weeks out: read the renewal terms against the expiring policy line by line, looking for new forms, changed definitions and sublimits. Ask about every change.
5. At binding: file the final policy, the carrier's written answers and the register as it stood on that date.
6. Mid-term: record any new tool or feature, and ask the broker whether it must be reported.

The value of the register shows at step three. Suppose the application asks whether the agency uses automated tools to give customers information about coverage. If the answer in your head is "no" but the register shows a website assistant that answers questions about opening hours and quote requests, you can answer precisely: it collects details and books callbacks, and it is scripted not to describe coverage. That is a true answer with its limits stated, and a better position than an unqualified "no" that a later claim contradicts.

Using AI to build the register, safely

AI can tidy the interview notes into a register, as long as the notes contain no client data. Paste only your own staff's descriptions of tools and uses.

Below are notes from interviews with five staff about the AI
tools they use at work. They contain no client information.
Create a table with columns: tool, who uses it, what for, what
kind of information goes in, where the output goes, who checks it
before it reaches a client. Where the notes do not say, write
"unknown - ask". Do not guess or fill gaps.

Illustrative output rows might read: "Meeting recorder | two account managers | notes from client review calls | spoken client information | file notes | unknown - ask". That last cell is the finding. The tool is in use, nobody has said who checks its output, and that is the question to resolve before you answer an underwriter.

Review the table in person with each user. People forget the free tool they use "just for emails", and that omission is the one that causes trouble later.

What a good answer to an underwriting question looks like

When an application or underwriter asks about AI use, aim for answers that are short, true and specific. Here is an illustrative before and after for the question "Does the agency use artificial intelligence in client-facing work, and what controls apply?"

The stronger answers are only strong if they are true. Do not describe controls you plan to introduce as if they exist, and do not describe a policy as "in place" until staff have been told about it. If the application has a signature, assume a claims handler may one day compare it with the register and the file notes.

Small agencies and the owner's own use

In a two-person agency the principal is often the heaviest user. The owner drafts proposals late at night with a chatbot, then sends them under their own licence. The practical risk is identical, but the review control is thin because the reviewer and the author are the same person. A workable rule: anything that states what a policy covers gets checked against the policy wording itself, not against the AI summary, and that check is noted in the file. It costs a few minutes per proposal and it is the cheapest control available. The same rule applies when you prepare for an account review: an AI summary is a starting point for questions, never a stand-in for the policy.

Three short scenarios to test your answers

Phone assistant mis-states a deductible. A caller asks about a deductible on an existing policy and the assistant guesses. Your policy tells it never to state coverage details; the transcript shows it did. This is a design failure, and the carrier question is whether the assistant's statement counts as agency work product.

Staff use a personal chatbot account. An account manager pastes a client's schedule of locations into a free tool. Nothing is wrong with the output, but the data left the agency. The question is cyber and privacy, not E&O, which is why the question list asks about both.

Renewal reminder with the wrong date. An AI-drafted reminder gives a renewal date a week late, and the policy lapses. This one looks like a plain human-style error, and most E&O policies were written for exactly that. What matters is whether the policy has an AI-related carve-out, and whether your own process required a person to verify dates.

Run your own agency through each scenario: which tool, which policy, which question, who checks. If you cannot answer one, that is the gap to take to your broker.

Questions agency principals ask about E&O and AI

Does a blanket AI exclusion mean I cannot use AI at all?

No. It means that a claim arising from AI use may not be covered under that policy, so you either avoid the exposed uses, negotiate narrower wording, or look for separate cover. Ask your broker to explain the exact wording before you decide, because the effect depends on how the exclusion is drafted.

Should I tell my carrier about every tool the staff use?

Tell them what the application and the policy conditions ask for, accurately. Start by finding out what staff actually use, including free tools on personal accounts, so that your answers are true. Ask your broker or coverage counsel what must be disclosed and when changes need to be reported.

Is a written AI policy required by carriers?

Not universally. Applications and underwriters may ask about AI governance, and a short written policy and a tool register are good evidence that you control the risk. Treat it as sensible practice and as material that supports a conversation with your carrier, not as a guarantee of cover.

What if my E&O policy is silent on AI?

Silence can mean cover applies as for any error, but it can also mean uncertainty. Ask the carrier to confirm in writing how AI-assisted work is treated. Keep their reply with the policy. Also check your general liability and cyber policies, because optional generative AI exclusions now exist for general liability and other policies can carry their own wording.

Who should I ask if the carrier's answer is unclear?

Your insurance broker or an adviser who specialises in professional liability cover. For disputes or a live claim, take legal advice. This tutorial helps you prepare the questions; it cannot interpret your policy for you.

Further reads

• AI Answering Service for Insurance Agents: Claims Calls and Quote Leads — AI answering for claims calls and quote leads.
• AI Insurance Agency Management System: What to Ask Before You Switch — Questions to ask before switching management systems.
• AI Insurance Lead Generation: Follow-Up Rules Before You Automate — Follow-up rules before you automate lead contact.
• Can AI Answer Calls and Qualify Leads for an Insurance Agency? — Whether AI can answer calls and qualify leads.
• AI for Insurance Compliance: Licence Tracking and Complaint Logs — Licence tracking and complaint logs with AI.
• Software for Insurance Agents and Brokers: Where AI Fits the Stack — Where AI fits in the agency software stack.
• What Does AI Cost a Small Insurance Brokerage in Year One? — What AI costs a small brokerage in its first year.
• AI Cross-Selling for Insurance Agencies: Account Review Prep, Not Advice — Account review prep without letting summaries replace policies.
• Professional Indemnity Renewal: Prepare the Proposal Form With AI — What insurers ask or may ask about AI at renewal, plus a claims summary, an AI-use register, a controls list and broker questions you can prepare.
• AI Insurance Quote Comparison for Brokers: Carrier Quotes to Proposal — A broker's workflow from three carrier quotes to a client proposal: normalise, flag gaps, draft the narrative, and check every cell before advising.
• AI Tools for Life Insurance Agents: Follow-Up Within the Rules — A rule-first method for AI follow-up in life insurance: guard-railed prompts, replacement flags, recording consent and a pre-send gate.
• ChatGPT for Insurance Agents: Ten Safe Uses and Three to Avoid — Ten uses an agency owner can approve for ChatGPT, and three to ban outright, with the client-data and policy rules that go with them.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: an independent agents' association summary of Verisk's ISO generative AI exclusions (CG 40 47, CG 40 48, CG 35 08, January 2026 edition, with the form definition); Hunton insurance recovery commentary on AI exclusions in professional lines (May 2025); your own policy wording. Check forms with your carrier.