Third-party service providers

Is My AI Tool a Third-Party Service Provider, and What Due Diligence Do I Owe?

Updated November 11, 2026

Short answer: If the tool stores, processes or can see your clients' information, the rules almost certainly treat it as a service provider. The NAIC's data security model law (#668), its safeguards model (#673) and New York's 23 NYCRR 500 each point an agency to choosing such a vendor with care and requiring it to protect the data; #668 and New York require it, and #673 gives it as its example of how to run the required program. That duty does not change because the AI is a feature inside software you already use. This page sets out the definitions, what each rule asks you to do, and a questionnaire you can send to any AI vendor.

What the definitions say

Each set of rules has its own term, but they describe the same thing: an outside business that gets access to your clients' information because it provides a service to you.

Rule Its term The definition, in its own words
NAIC #668 (section 3P) Third-Party Service Provider "a Person, not otherwise defined as a Licensee, that contracts with a Licensee to maintain, process, store or otherwise is permitted access to Nonpublic Information through its provision of services to the Licensee"
NAIC #673 (section 2E) Service provider "a person that maintains, processes or otherwise is permitted access to customer information through its provision of services directly to the licensee"
23 NYCRR 500.11 (New York) Third-party service provider The policy must cover "information systems and nonpublic information that are accessible to, or held by, third-party service providers"
NAIC AI bulletin (insurers only) Third Party "an organization other than the Insurer that provides services, data, or other resources related to AI"

On the plain words of #668, an AI vendor that processes your clients' information "through its provision of services" to your agency is a third-party service provider. The AI bulletin's definition is broader still, but it is written for insurers, so it reaches an agency only through a carrier's own program.

Sources: NAIC #668, NAIC #673, 23 NYCRR 500 (DFS official copy) and the NAIC AI bulletin.

What due diligence the rules ask for

Under #668, section 4F has two parts: "(1) A Licensee shall exercise due diligence in selecting its Third-Party Service Provider; and (2) A Licensee shall require a Third-Party Service Provider to implement appropriate administrative, technical, and physical measures to protect and secure the Information Systems and Nonpublic Information that are accessible to, or held by, the Third-Party Service Provider." The written program it sits in is sized to the agency, "including its use of Third-Party Service Providers" (section 4A).

Under #673, section 8 gives, as its example of how to run the required program, a licensee that exercises "appropriate due diligence in selecting its service providers", requires them "to implement appropriate measures designed to meet the objectives of this regulation", and, "where indicated by the licensee's risk assessment", takes "appropriate steps to confirm that its service providers have satisfied these obligations." That last part is about checking, not only asking. The model calls sections 6 to 9 "examples of methods of implementation", so check how your state's version words them.

Under 23 NYCRR 500.11, a New York covered entity needs "written policies and procedures designed to ensure the security of information systems and nonpublic information that are accessible to, or held by, third-party service providers." The regulation says the policy covers risk assessment of vendors, minimum practices they must meet, due diligence and periodic assessment, and guidelines on access controls, multi-factor authentication, encryption, breach notice and representations and warranties. Among them is "notice to be provided to the covered entity in the event of a cybersecurity event directly impacting ... the covered entity's nonpublic information being held by the third-party service provider". Source: the DFS's official copy of Part 500. Even small agencies with the limited exemption must keep this policy.

Under the NAIC AI bulletin, insurers are asked to cover "Due diligence and the methods employed by the Insurer to assess the third party and its data or AI Systems", and, "Where appropriate and available", contract terms that "Provide audit rights and/or entitle the Insurer to receive audit reports by qualified auditing entities" and "Require the third party to cooperate with the Insurer with regard to regulatory inquiries". These are insurer duties. If you sell for a carrier in an adopting state, they show what the carrier may ask of its own vendors, and possibly of you.

"It's inside our agency management system" does not change the answer

Many agencies first meet AI as a new feature inside software they already pay for, or as a partner add-on that connects to it. Our reading is that this does not change the analysis. An AI feature that handles client information, whether the software vendor runs it or a partner does, is a third-party service provider like any other. The vendor relationship you already have is a good place to ask, but it does not replace the due diligence. The software vendor questions are written for exactly that case.

A questionnaire for any AI vendor

These are the questions that map to the duties above. Send them in writing, and keep the answers with your information security program.

Training and use of your data

  1. Do you, or any AI model provider you use, train any model on our clients' information? Is that the default, and can it be changed by an opt-in?
  2. Do you run our data through automated classifiers or safety tools, and what happens to data they flag?

What is kept, and for how long

  1. Do you store our prompts, files and the AI's answers? For how long by default, and can our admin set a shorter period?
  2. When we delete something, how long until it is gone from all your systems, including backups? What exceptions apply?
  3. Is zero data retention available on the plan we would buy, or only on a different product such as an API?

Who can see it

  1. Which of your employees or contractors can access our stored content, and for what reasons?
  2. Which other companies, such as model providers or hosting partners, receive our data, and what are their terms?
  3. Where is our data stored and processed? Does any of it leave the United States?

When something goes wrong

  1. How quickly will you tell us about a cybersecurity event that affects our clients' information? Our #668 clock can start the day after you tell us.
  2. What do you do when you receive a subpoena or other legal demand for our data, and will you tell us?

Contracts and proof

  1. Will you sign a data processing addendum, and will you put your safeguards in the contract?
  2. Do you have independent audit reports, such as a SOC report, and will you share them?
  3. If we write health or life business, will you sign a business associate agreement where one is needed?

Records

  1. If we must keep a record of what was sent to a client, where does that record live: with you, or in our own systems?

What good answers look like, from two common plans

Two published examples show how the answers vary. ChatGPT Business says "By default, we do not use your business data for training our models", lets admins "control how long your data is retained", and names the OpenAI staff and contractors who may access stored chats (OpenAI). Microsoft 365 Copilot says prompts and responses "aren't used to train foundation LLMs", and stores them as your organization's Microsoft 365 content, with retention set by your admin in Microsoft Purview (Microsoft Learn). Neither answer is right or wrong in general. Each tells you what your program has to cover. The small agency options page compares these and others.

Keep a record of what you found

Due diligence you cannot show is hard to defend. For each AI tool, keep the vendor's answers, the date you got them, the plan and settings you chose, and the contract terms you signed. Review them when the vendor changes its terms, and at least when you review your information security program.

This guide is general information, not legal or compliance advice. Your state's law and your own counsel decide what your agency's program must include.

About Faradex. Faradex is a private AI workspace for small and mid-sized firms that handle confidential client work. Each firm gets its own Workspace on dedicated infrastructure in the United States. Files and conversations are held in memory and are gone when the Workspace is reset. AI model: Currently Opus via AWS Bedrock, more information is available on our website. AWS does not store prompts or responses, and model providers have no access to them. Pricing: from $120 a month, no seat minimum; see faradex.ai/pricing.