AI for insurance agents: how to use ChatGPT and other AI tools without creating a compliance problem
By Dan Blomquist, Aegis AI Advisory · Published November 10, 2026 · Sources checked November 10, 2026
Short answer: insurance agents can use ChatGPT, Copilot, Gemini and similar tools for a lot of everyday work. Three rules keep it out of trouble:
- No client nonpublic information in personal AI accounts. Names with Social Security, driver’s license or account numbers, and health details, stay out of Free, Plus or Pro chatbots.
- A licensed person checks anything a client will rely on, such as coverage explanations, quote comparisons and claim letters, before it goes out.
- No AI-voiced calls without consent. The FCC treats AI-generated voices as “artificial” under the TCPA.
The NAIC’s AI model bulletin is written for insurers, not agencies, but carriers that follow it are asked to look at AI across sales and distribution. A short written AI policy is how an agency shows it has these rules.
Most search results for “AI for insurance agents” are tool lists. This page is about the part those lists skip: what the rules say about using these tools with client data, and what your agency’s written policy should say. If you want the basics of any staff AI policy first, start with what an AI acceptable-use policy should contain.
Which rules reach an agency, and which don’t
Three sets of rules come up when an agency adopts AI. They apply to different people, and it helps to keep them straight.
| Rule | Written for | What it means for an agency |
|---|---|---|
| NAIC Model Bulletin: Use of AI Systems by Insurers (adopted December 4, 2023) | Insurers holding a certificate of authority in an adopting state. | Not addressed to agencies. But it asks each insurer’s written AIS program to cover “sales and distribution” and AI systems or data from third parties, with due diligence and, where available, audit and cooperation terms in contracts. Expect it to show up in carrier questions and agreements. |
| State insurance data security law (NAIC Model #668; in Michigan, MCL 500.550–500.565) | “Licensees,” which in Michigan’s text means a licensed insurer or producer and others licensed under the Insurance Code. | Applies to your agency directly. It defines the client data you must protect, covers vendors that can access it, and sets a 10-business-day deadline to notify the regulator of certain cybersecurity events. |
| FCC Declaratory Ruling FCC 24-17 (February 2024) | Anyone placing calls covered by the TCPA. | AI-generated voices count as “artificial or prerecorded voice,” so calls using them need the called party’s prior express consent. Relevant to AI dialers and voice agents for lead follow-up. |
The NAIC’s implementation map (status as of November 3, 2026) lists 27 states, the District of Columbia and Puerto Rico as having adopted the AI bulletin, and California, Colorado, New York and Texas as having their own insurance-specific AI rules or guidance. The NAIC’s data security model brief lists 28 jurisdictions that have implemented Model #668. Small-agency exemptions differ: the model exempts licensees with fewer than 10 employees from the written program, while Michigan’s threshold is fewer than 25. Check your own state’s text.
Everyday agency AI tasks: fine, careful, or off-limits
This is the table most agency policies need and few have. It assumes staff are on a personal or free AI account unless your policy says otherwise.
| Task | Call | Why |
|---|---|---|
| Draft a renewal reminder, newsletter or social post with no client details | Fine | No nonpublic information. A person still checks claims about coverage or price before it is published. |
| Rewrite a coverage explanation in plain English | Careful | Useful, but AI can misstate what a policy covers. A licensed agent checks it against the actual policy form before a client sees it. |
| Summarize a carrier bulletin, underwriting guide or a blank policy form | Fine | Usually public or business information. Check your carrier agreement for limits on uploading carrier materials. |
| Paste a client’s application, driver’s license or loss run into a chatbot | Off-limits | A name plus a driver’s license, Social Security or account number is nonpublic information under MCL 500.553(i). Only an approved business tool with vendor terms you have reviewed. |
| Life or health underwriting notes, medical records, claim details | Off-limits | Health information that identifies a consumer is also nonpublic information under Michigan’s definition, with no other data element needed. |
| AI notetaker on client calls or meetings | Careful | The vendor holds recordings of nonpublic information, so treat it as a service provider to review. Recording-consent rules vary by state. |
| AI voice agent or AI-voiced voicemail to leads | Off-limits without consent | Prior express consent required under the FCC’s 2024 ruling; marketing calls carry stricter consent rules. |
| Website chatbot or AI receptionist that takes quote requests | Careful | Prospects will type in personal details. Review the vendor’s data terms and tell visitors they are talking to AI. Our look at whether an AI receptionist is worth it covers the operating side. |
A starting point, not legal advice. Your E&O carrier, your carrier agreements and your state’s rules may be stricter.
Why the account type matters
OpenAI’s help center says that for its services for individuals (Free, Plus and Pro) it may use your content to train its models unless you turn off “Improve the model for everyone” in Data controls, and that thumbs-up or thumbs-down feedback can send the whole conversation into training even after opting out. For ChatGPT Business, Enterprise and the API, OpenAI says it does not use inputs or outputs to train its models by default, and its enterprise privacy page says it can sign a data processing addendum for those plans. That is why an agency policy names the approved tools and plans instead of just saying “be careful with AI.” The same check applies to Copilot, Gemini and any AI feature inside your agency management system: find out which plan your staff actually sign in with and what its terms say.
Under Michigan’s law, an outside vendor that is permitted access to nonpublic information through its services to you is a “third-party service provider” (MCL 500.553(l)). Licensees covered by the written-program requirement must use due diligence in selecting those providers and require them to protect the data (MCL 500.555(6)). An AI tool that holds your clients’ applications is that kind of vendor, whatever its marketing page calls it.
Notes for Michigan agencies
- Michigan adopted the AI bulletin. DIFS issued Bulletin 2024-20-INS on August 7, 2024. It applies to insurers, nonprofit health service plans, HMOs and dental plan organizations, and ties AI-supported decisions to the Unfair and Prohibited Trade Practices and Frauds Act (MCL 500.2001–500.2093) and to MCL 500.2403’s rule that property and casualty rates not be excessive, inadequate or unfairly discriminatory.
- Small agencies are exempt from the written program, not from notice. Under MCL 500.565, a licensee with fewer than 25 employees, counting independent contractors, is exempt from the information security program in section 555. DIFS points out that the exemptions do not apply to the cybersecurity event notification rules in sections 559 and 561. If you grow to 25, you have 180 days to comply.
- The 10-business-day clock. A producer whose home state is Michigan must notify the DIFS director within 10 business days after determining that a cybersecurity event involving nonpublic information has occurred, when it is reasonably likely to materially harm a Michigan consumer or a material part of the agency’s operations (MCL 500.559(1)). A client file uploaded to the wrong AI account belongs in your incident process, so someone can decide quickly whether it qualifies.
- Captive agents. An agent who is also a licensee and is covered by the insurer’s information security program is exempt from building a separate one (MCL 500.565(3)). Ask your carrier whether its program, and its AI rules, cover the tools you use.
Many independent agencies, from metro Detroit to Marquette and the rest of the Upper Peninsula, run on a handful of people and an agency management system plus Microsoft 365 or Google Workspace. That is exactly the size where a one-page AI rule set does more than a binder.
What goes in an insurance agency AI policy
- Owner: one named person (often the agency principal) who approves tools and answers questions.
- Approved tools list: the tool, the plan, whether it may hold client data, and who signed off on its terms.
- Client data rule: no nonpublic information in any tool that is not on the approved list; no personal accounts for client work.
- Licensed review: a licensed agent reviews any AI-assisted coverage explanation, comparison, recommendation or claim communication before it reaches a client.
- Marketing and calls: AI-written ads and posts are checked for accuracy; no AI-voiced calls or voicemails without documented consent.
- Client-facing AI: chatbots and AI receptionists tell people they are AI and hand off to a person on request.
- Carrier requirements: where a carrier agreement or carrier AI program sets rules, those win.
- Incidents: what to do if client data goes into an unapproved tool, and who decides whether it must be reported.
- Training and sign-off when someone starts and whenever the approved list changes.
The general sections (purpose, prohibited uses, annual review) follow the same pattern as any firm’s policy; compare the free sample policy. Agencies that also prepare taxes or offer financial services will recognize the client-data rules from our CPA firm guide, and an agency that sells health plans should read the HIPAA guide for how business associate agreements work.
Sample clauses for an agency
Excerpt for illustration only. Not legal advice. [Agency] and [Principal] are placeholders.
3. Client information
Staff must not enter a client’s or prospect’s nonpublic information, including Social Security, driver’s license or financial account numbers, health information, or a full application, loss run or claim file, into any AI tool unless the tool and plan are on [Agency]’s Approved AI Tools list for client data. Personal or free AI accounts may never be used for client work.
4. Licensed review
AI output is a draft. A licensed agent must check any AI-assisted coverage explanation, quote comparison, recommendation or claim-related message against the actual policy, quote or carrier guidance before it is sent to a client. The licensed agent, not the tool, is responsible for what the client is told.
6. Calls and messages
No outbound call or voicemail that uses an AI-generated or cloned voice may be placed without the recipient’s documented prior consent, confirmed by [Principal].
8. Mistakes
If client information is entered into a tool that is not approved, tell [Principal] the same day. [Principal] will decide whether it is a cybersecurity event that must be reported to the Department of Insurance and Financial Services or anyone else.
Start this week
- Ask everyone, including part-time CSRs and producers who work from home, which AI tools they use for agency work. No penalty for honest answers; our shadow AI guide explains why this comes first.
- For each tool, write down the plan and whether it trains on your data. Anything on a personal account is “no client data” from today.
- Read the AI or technology sections of your carrier agreements, and ask your E&O carrier whether it has AI questions on its application.
- Adopt a short written policy and get signed acknowledgments. The 90-day AI governance plan shows how to sequence the rest.
What isn’t settled
The NAIC bulletin is principles-based, and each state’s version can differ from the model; the NAIC says its map does not judge whether a state’s text matches. How carriers will apply their AIS programs to independent agencies is still taking shape, and it will vary by carrier. Consent rules for AI-voiced calls depend on the type of call and on state law as well as the TCPA. Vendors also change which plans train on your data and what their terms cover, so the account-type section above is a snapshot. A written policy makes your choices explicit and easy to update; it does not replace your carrier’s requirements, your E&O carrier’s guidance or an attorney.
Get a policy drafted for your agency
Answer 10 questions, choose Insurance, and the generator emails you an AI acceptable-use policy that addresses state insurance data security law, the NAIC model bulletin on AI, unfair discrimination and agent licensing responsibilities, using the tools and state you enter.
- $49 Policy: AI acceptable-use policy (Word + web version) and a one-page staff checklist.
- $149 Pro Pack: adds an AI vendor risk register and a staff training sign-off sheet.
Not ready? Read the free sample policy first, or fill in our $9 DIY Starter Kit.
Disclosure: Aegis AI Advisory sells this generator. It produces a template, not legal advice, and it does not review carrier agreements or vendor contracts. It is not the right fit if your carrier or an attorney has already given you an AI policy you must follow, or if you are an insurer building an AIS program under the NAIC bulletin. For an agency that wants an AI usage inventory, vendor review and staff training, see the AI governance track or book a free AI Operations Briefing.
Frequently asked questions
Can insurance agents use ChatGPT?
Yes, for work that does not involve a client's protected information: drafting a renewal reminder, rewriting a coverage explanation in plain English, summarizing a carrier bulletin, or brainstorming a newsletter. The line is client data. Michigan's insurance data security law treats a consumer's name combined with a Social Security number, driver's license number or financial account number, and health information that identifies a consumer, as nonpublic information. That should not go into a personal Free, Plus or Pro ChatGPT account, where OpenAI says it may use conversations to train its models unless the user opts out. OpenAI says it does not train on ChatGPT Business or Enterprise data by default. Either way, the agent is still responsible for checking what the AI wrote before a client sees it.
Does the NAIC AI model bulletin apply to insurance agencies?
Not directly. The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted December 4, 2023, is addressed to insurers that hold certificates of authority, and it expects each insurer to keep a written AI Systems (AIS) program. But it asks that program to cover the whole insurance life cycle, including marketing and sales and distribution, and AI systems or data that come from third parties. So an agency's AI use can come up in carrier due diligence or agency agreements, even though the bulletin does not regulate the agency itself.
Has Michigan adopted the NAIC AI bulletin?
Yes. Michigan DIFS issued Bulletin 2024-20-INS, Use of Artificial Intelligence Systems by Insurers, on August 7, 2024. It follows the NAIC model and reminds insurers, nonprofit health service plans, HMOs and dental plan organizations that AI-supported decisions must comply with Michigan insurance law, including the Unfair and Prohibited Trade Practices and Frauds Act (MCL 500.2001 to 500.2093) and the rule that property and casualty rates not be excessive, inadequate or unfairly discriminatory (MCL 500.2403).
How many states have adopted the NAIC AI model bulletin?
The NAIC's implementation map, with status as of November 3, 2026, lists 27 states plus the District of Columbia and Puerto Rico as having adopted it. It separately lists California, Colorado, New York and Texas as having their own insurance-specific AI regulation or guidance. The NAIC notes the map does not determine whether each state's version contains every element of the model, so read your own state's bulletin.
Does Michigan's insurance data security law apply to a small agency?
Partly. Chapter 5A of the Michigan Insurance Code (MCL 500.550 to 500.565) applies to licensees, which includes licensed producers. A licensee with fewer than 25 employees, counting independent contractors, is exempt from the written information security program requirement in MCL 500.555. DIFS notes that the exemption does not reach the cybersecurity event notification rules in MCL 500.559 and 500.561. A producer whose home state is Michigan must notify the DIFS director within 10 business days of determining that a qualifying cybersecurity event involving nonpublic information occurred.
Can my agency use an AI voice agent to call leads?
Only with the right consent. In a February 2024 declaratory ruling (FCC 24-17), the FCC confirmed that the Telephone Consumer Protection Act's restrictions on an "artificial or prerecorded voice" cover current AI technologies that generate human voices, so calls using them require the prior express consent of the called party. An agency policy should say that no AI-voiced outbound call or voicemail goes out without documented consent, and that a person, not the tool, owns that decision.
What does the Aegis AI policy generator produce for an insurance agency?
Choose Insurance as your industry and answer 10 questions. The $49 Policy is an AI acceptable-use policy in an editable Word file and a web version, plus a one-page staff checklist. For insurance it is written to address state insurance data security law (the NAIC Insurance Data Security Model Law and state versions such as Michigan's), the NAIC Model Bulletin on AI use by insurers, unfair discrimination, and the agent's licensing responsibilities, using the tools and state you enter. The $149 Pro Pack adds an AI vendor risk register and a staff training sign-off sheet. It is a template, not legal advice.
Sources
- NAIC Model Bulletin: Use of Artificial Intelligence Systems by Insurers (adopted December 4, 2023)
- NAIC, Implementation of NAIC Model Bulletin: Use of AI Systems by Insurers (status as of November 3, 2026)
- Michigan DIFS Bulletin 2024-20-INS, Use of Artificial Intelligence Systems by Insurers (August 7, 2024)
- Michigan Insurance Code, Chapter 5A Data Security (MCL 500.550–500.565) and MCL 500.565
- Michigan DIFS, Michigan Insurance Data Security Law
- NAIC, Insurance Data Security Model Law (#668) brief
- FCC, Declaratory Ruling FCC 24-17: TCPA applies to AI technologies that generate human voices
- OpenAI: How your data is used to improve model performance, Enterprise privacy
Also free: check whether ChatGPT and Google’s AI can find your agency → · More guides on the resources page.