Audit Rights in Insurer AI Vendor Contracts Under the NAIC AI Model Bulletin

What the bulletin expects of third-party AI contracts

The NAIC adopted the Model Bulletin on the Use of AI Systems by Insurers on December 4, 2023. It is a model bulletin rather than state law, so the obligations an insurer faces depend on how each state has adopted it. Its core message is that governance, risk management controls and internal audit functions play a central role in reducing the risk that AI-driven decisions violate unfair trade practice laws. The insurer's AIS Program is expected to address all three.

For third parties, the bulletin's language is specific. It refers to including contract terms that provide audit rights and/or entitle the insurer to receive audit reports from qualified auditing entities, where appropriate and available. It also refers to the performance of contractual rights regarding audits and/or other activities to confirm the third party's compliance with contractual and, where applicable, regulatory requirements.

Two points follow. First, a clause alone is not enough: the bulletin contemplates that audits or confirmation activities are actually carried out. Second, the insurer remains the party answering to the Department. The bulletin describes the information a Department may request during an investigation or examination, and that list includes vendor contracts, records of audits and confirmation processes, and documentation on validation, testing and auditing, including evaluation of model drift.

Where contractual rights fall short

Every mechanism the bulletin names for vendor oversight runs through the vendor or an outside auditor: an audit right, or a report from a qualified auditing entity. If the vendor is slow, restricts scope, or treats a model as proprietary, the right may produce little usable evidence. Contract language also says nothing about what the AI actually did in a particular interaction at a particular time.

The qualifier "where appropriate and available" matters here. It acknowledges that an insurer will not always secure a full audit right. A reasonable reading is that when a right is absent, the insurer should document why and record what alternative confirmation activities it performed. That is an inference from the text, not a stated rule, but it fits the bulletin's emphasis on demonstrable oversight.

The stricter July 2023 exposure draft is also worth noting. It would have required contract terms entitling the insurer to audit the vendor for compliance, without the "where appropriate and available" condition. It was superseded, so the adopted text is the reference. The draft language can still be a useful opening position in negotiation.

Vendor-dependent evidence versus insurer-controlled evidence

Separating these two categories shows where an insurer is exposed. A contract can improve the first category. Only the insurer's own practices can build the second.

• Audit reports and audit rights: Vendor or auditor dependent. Obtained through the contract; scope and timing depend on the vendor.
• Validation, testing and drift documentation: Largely vendor dependent. The insurer should confirm in advance that it can be supplied and produced on request.
• Records that audits or confirmation activities were performed: Insurer controlled. The insurer creates and retains them.
• Pre-acquisition and pre-use diligence, ongoing monitoring and oversight records: Insurer controlled. Retain them internally rather than relying on the vendor to reconstruct them.
• Evidence of what the AI was permitted to do and did in each interaction: Insurer controlled only if captured on the insurer's side at runtime. Vendor records alone cannot show this independently.

How to evaluate a draft or existing contract

Work through the contract in a fixed order so compliance, legal and procurement reach the same conclusions.

• Map each clause to the bulletin's named topics: audit rights or audit reports, representations and warranties, data sourcing, data security and privacy, IP, confidentiality and disclosures, and cooperation with regulators.
• Mark each right as exercisable by the insurer or dependent on the vendor's goodwill, and note any scope limits on proprietary models and data.
• Test the evidence path: ask whether the vendor could deliver validation, testing and drift documentation on a regulator's timeline.
• Where a right is missing or limited, record the rationale and the alternative confirmation activity the insurer will perform.
• Schedule and document the audits or confirmation activities themselves, since the bulletin refers to their performance and not only to the contract term.
• Identify which oversight evidence the insurer will generate and retain itself, and assign an owner.

Building evidence the insurer owns

The practical answer to vendor dependence is to hold independent records. Some are procedural: diligence files, audit and confirmation records, and monitoring reviews. Others are technical: a record of what the AI was permitted to do and what it did, generated at the time of the interaction and retained under the insurer's control.

Trussed enforces policy in the execution path for models, agents, tools and MCP servers, applying allow, redact, block, rate limit or escalate for human approval to individual interactions and tool calls. Enforcement is at runtime, not reviewed after the fact, and it produces audit-ready evidence mapped to NAIC, among other regimes. The platform can be self-managed in the insurer's own cloud, with configurable retention and data residency.

This does not replace contractual audit rights, and it is not a substitute for vendor validation, testing or drift documentation. It narrows the gap those rights leave: when a regulator asks what controls applied to an AI interaction, the insurer can answer from its own records without waiting on the vendor.