NAIC AI Bulletin and Insurance Chatbots: Program Requirements and Runtime Enforcement
The NAIC AI Model Bulletin expects insurers to maintain a written AIS Program covering governance, risk management controls and internal audit, and it holds insurers responsible for insurance-law compliance when they use AI such as chatbots. It does not specify how an individual chatbot interaction is controlled or evidenced. Carriers and brokerages must translate the program language into runtime controls (allow, redact, block, escalate) and keep evidence that those controls operated. Status below is as of October 2026.
Evaluation criteria for runtime enforcement and evidence
- Does the control act in the chatbot's execution path, or only review activity afterwards?
- Can each interaction-level action be exported as evidence and mapped to AIS Program elements: governance, risk management controls and internal audit?
- How does it behave with third-party models and tools, and what does it record about calls to them?
- How are redaction, prohibited-output and human-approval policies defined, versioned and changed?
- Can you set retention and data residency to match your obligations?
- Which examination evidence does it produce, and which remains your program-level documentation?
Mapping chatbot risk points to program expectations and runtime controls
Customer data exposure
Program element: risk management controls. Runtime control: redact personal information before it reaches a model or tool, block where policy prohibits the transfer, and record each action.
Unfair or unsupported statements
Program element: governance and insurance-law compliance. Runtime control: block prohibited or unsupported outputs, escalate defined categories for human approval, and record the decision and approver.
Third-party models and tools
Program element: oversight of vendor-supplied systems, including contract terms. Runtime control: apply the same policy whichever model is called, allow only approved tools, and record calls to them.
What the bulletin sets at the program level
NAIC adopted the Model Bulletin on the Use of Artificial Intelligence Systems by Insurers in December 2023, aligned with the NAIC Principles of Artificial Intelligence. It reminds insurers that decisions or actions made or supported by AI must comply with all applicable insurance laws and regulations. It also sets expectations for how insurers govern AI use and advises them of the type of information a department may request during an investigation or examination.
NAIC's own description of insurer AI use includes chatbots that answer common customer questions, provide basic information and assist with simple transactions. The same material states that insurers remain responsible for compliance when they use AI. A chatbot built on a vendor's model does not move that responsibility.
The central structure is the AIS Program, a written program for the responsible use of AI Systems that make or support decisions related to regulated insurance practices. It should address governance, risk management controls and internal audit functions. Whether a given chatbot falls inside the program depends on whether it makes or supports decisions tied to regulated insurance practices. Inventory both customer-facing and internal chatbots, and record that determination for each.
Where adoption and related regimes stand as of October 2026
The model is not self-executing. It takes effect only where a state issues its own bulletin or guidance, and operative dates vary. Illinois Company Bulletin 2024-08 (March 2024), New Jersey Insurance Bulletin No. 25-03 (February 2025) and Hawaii Memorandum No. 2025-13A (December 2025) are examples of state action. NAIC's adoption map states that it does not determine whether a state's action contains all elements of the model, so review the text of each state where you operate.
NYDFS Insurance Circular Letter No. 7 (July 11, 2024) addresses AI systems in underwriting and pricing. It expects insurers to incorporate certain terms into third-party vendor contracts where appropriate, drawing on the NAIC Model Bulletin. Its stated scope is underwriting and pricing, so it is not a chatbot rule, but the vendor-contract expectation is relevant to chatbots built on third-party models.
Colorado has changed. The Colorado Attorney General reports that Senate Bill 26-189, signed in May 2026, repeals and reenacts the earlier provisions of SB 24-205 with new requirements on automated decision-making technology in consequential decisions. Do not use SB 24-205 as a current baseline. Reassess any Colorado exposure against the new law.
Program-level duties versus interaction-level enforcement
A written AIS Program is necessary but describes intent. It contains policies, assigned owners, risk assessments, an internal audit function and vendor contract terms. An examiner who asks how the program works for a chatbot will also want to know what a specific conversation was allowed to do and what happened. That is an interaction-level question: which personal information was redacted, which output was blocked, which request went to a human and who approved it, and which tool calls were made.
The bulletin material does not describe how those events must be controlled or recorded, so the translation falls to the compliance team. Two design choices matter. First, controls that operate in the execution path enforce policy at runtime, while log review only identifies problems after the fact. Second, evidence is most reliable when the control that acted also produces the record, rather than being reconstructed from separate systems.
There are tradeoffs. Strict blocking rules can stop legitimate answers. Escalation for human approval adds queue volume and delay. Policies change, so each change needs versioning and an owner. Retention and data residency settings for the evidence itself also need a decision. None of this replaces the written program. Runtime evidence supports the program elements, and the program remains the carrier's own document.
Two layers of chatbot compliance
Label: Written AIS Program
Description: Governance, risk management controls and internal audit, owned by the carrier or brokerage.
Label: Runtime enforcement
Description: Policy applied in the chatbot's execution path: allow, redact, block, escalate for human approval.
Label: Interaction-level evidence
Description: A record of what each control did, mapped back to program elements.
Review your chatbot controls against the AIS Program
Walk through how policy enforcement in the execution path produces audit-ready evidence for insurer chatbots.
Request a Demo