Short answer: Ask the same questions you would ask any AI vendor, because the rules treat an AI feature inside your agency management system, or a partner add-on that connects to it, as a service provider like any other. Ask what it sends where, what it keeps, who can see it, whether anything trains on it, and how fast you will hear about a breach. Get the answers in writing. This page is a list of questions, grouped by topic, with a line on why each one matters. It makes no claims about any vendor's terms.
Why the software you already use needs its own questions
AI is arriving in agencies through two doors. One is a general tool that staff sign up for, such as a chat assistant. The other is a new feature in software the agency already runs, such as its agency management system, its email or its document tools, or a partner product that plugs into them.
The second door is easy to overlook, because the vendor is already trusted. But the NAIC's data security model law (#668) asks a licensee to "exercise due diligence in selecting its Third-Party Service Provider" and to require it to protect the data it holds (section 4F). New York's 23 NYCRR 500.11 asks the same, and the safeguards model (#673, section 8) gives it as its example of how to run the required program. Our reading is that an AI feature that handles client information is covered by those duties whether it comes from a new vendor, an existing one, or a partner your existing vendor works with (how the rules define AI vendors).
We have not verified any agency management system vendor's AI data terms, so this page makes no claims about them. The questions below are how you find out, from your own account contact, in writing.
Questions about what the AI feature does
- Which features use AI, and which of them handle client information? Some features may only work on your agency's own settings or templates; others read policies, emails or client files.
- Is the AI built by you, or does it come from another company? Name the model provider, and any partner whose product does the work.
- When a feature runs, what data leaves your system, and where does it go? This tells you whether a new service provider is involved at all.
- Can we turn each AI feature on or off, for the whole agency and for each user?
Questions about training
- Is any of our clients' information used to train or improve a model, yours or anyone else's? Ask whether the answer is a default, and whether anyone at your agency could opt in by mistake.
- Does the same answer hold for every partner and model provider involved? A no-training promise from your vendor may not cover a partner with its own terms.
Questions about what is kept, and for how long
- Do you store our prompts, the documents the feature reads, and the answers it gives? For how long?
- Can our admin set or shorten that period?
- When we delete something, how long until it is gone from every system, including backups, logs and partners' systems?
- Is any of it kept for abuse monitoring or safety review? Who reviews it, and for how long?
Questions about who can see it
- Which of your employees or contractors can access our clients' information through the AI feature, and for what reasons?
- Where is the data stored and processed? Does any of it leave the United States?
- What do you do when you receive a subpoena or other legal demand for our data, and will you tell us?
Questions about security incidents
- How quickly will you tell us about a cybersecurity event that affects our clients' information? Under #668, a vendor's event is handled as the agency's own (section 6D), and our #668 clock can start the day after the vendor tells us. In New York, the 72-hour notice in 23 NYCRR 500.17 covers incidents at "a third-party service provider".
- Does that notice duty also bind your AI partners and model providers, and will it reach us in time?
Questions about contracts and proof
- Will you put the AI feature's safeguards, retention and training terms in our contract, or in a data processing addendum? #668 asks the agency to "require" the vendor's safeguards; a contract term is the usual way to show you did.
- Do you have an independent audit report, such as a SOC report, that covers the AI feature? Will you share it?
- If we write health or life business, does the AI feature handle health information, and will you sign a business associate agreement where one is needed? The safeguards model #673 covers "nonpublic personal health information" as well as financial information.
- If our carrier asks how we use AI, can you give us what we need to answer? The NAIC's AI bulletin asks insurers to look at audit rights and cooperation from third parties, and a carrier may pass those expectations on.
Questions about records
- Where does a record of what the AI produced live? If staff use it to draft something your agency must keep, such as a coverage explanation sent to a client, find out whether the record stays in the vendor's system, in your agency management system, or nowhere unless someone saves it.
How to use the answers
- Send the questions in writing, and ask for written answers, with the date.
- Keep the answers in your information security program file, next to the vendor's contract. That is your record of due diligence.
- Decide feature by feature. A vendor may have good answers for one AI feature and none yet for another. You can turn on the first and wait on the second.
- Ask again when the terms change. AI features change quickly. Review the answers when the vendor announces a new feature, and when you review your program.
The vendor questionnaire covers general AI tools, and the small agency options page shows how the published plans answer these questions.
This guide is general information, not legal or compliance advice. Your state's law and your own counsel decide what your agency must ask and keep.