Exam pack · NAIC AI Risk Evaluation Supplement v6.0 (exposure draft, 16 Oct 2026); v7.0 pending adoption vote, Fall National Meeting
Bluebonnet Mutual Insurance Company (fictional)
NAIC code DEMO-10101 · period 2026-07-02 to 2026-10-31 · all states · built 2026-10-30 11:45
- Pack identifier
- pack_e4072ef8b88d
- Pack hash (on ledger #1375)
- c56fad6b39de203033bd60e843014f878c2cefc5e1ac9c22ef9f6c754297a454
- Engine
- regbench-engine/0.4.0 · rules 2026.10+dbe5ddf6 · corpus 2026-10-30-5f07c521
- Totals
- 577 assessments · 547 case files · 4 high-risk models
- Confidentiality
- Produced for the requesting state's examination or inquiry; to be held confidential under that state's authority. The carrier may assert trade-secret protection over model documentation it attaches.
Exhibit A: Part One - Quantify Regulated Entity's Use of AI Systems
- Materiality threshold
- Models touching more than 2% of annual written premium or any direct consumer outcome; rationale: FCEH financial-position test plus consumer-impact floor.
- Lines
- auto, home
- Window for column 3
- 12 months
| (1) Operation or program area | (2) Systems in use | (3) Implemented in past 12 months | (4) Use case · direct consumer impact · material financial impact | (5) GenAI | (6) Agentic | (7) GLM | (8) Other AI/ML |
|---|---|---|---|---|---|---|---|
| Marketing | 1 | 1 | agent-assist: Generative AI assistant drafting service responses for human agents; no decision authority. Observed 30 decisions in the period, 0 adverse to the consumer. direct impact: False financial: immaterial | 1 | |||
| Premium Quotes & Discounts | 0 | 0 | |||||
| Underwriting/Eligibility | 1 | 0 | uw-triage: Scores new auto and home applications for eligibility and refers borderline risks to an underwriter. Observed 150 decisions in the period, 26 adverse to the consumer. direct impact: True financial: material | 1 | |||
| Ratemaking/Rate Classification/Schedule Rating/Premium Audits | 2 | 0 | price-opt: Recommends renewal premium within filed rating plan bounds using telematics, credit-based insurance score and territory. Observed 168 decisions in the period, 45 adverse to the consumer. direct impact: True financial: material uw-triage: Scores new auto and home applications for eligibility and refers borderline risks to an underwriter. Observed 150 decisions in the period, 26 adverse to the consumer. direct impact: True financial: material | 1 | 1 | ||
| Claims/Adjudication | 2 | 1 | claims-fastpath: Observed in the decision stream only (not in the declared inventory): claims-fastpath. Observed 109 decisions in the period, 10 adverse to the consumer. direct impact: True financial: unknown fnol-fraud: Vendor model scoring first notice of loss for fraud indicators; high scores route to SIU and can hold payment. Observed 120 decisions in the period, 6 adverse to the consumer. direct impact: True financial: moderate | 2 | |||
| Customer Service | 1 | 1 | agent-assist: Generative AI assistant drafting service responses for human agents; no decision authority. Observed 30 decisions in the period, 0 adverse to the consumer. direct impact: False financial: immaterial | 1 | |||
| Utilization Management/Utilization Review/Prior Authorization/Level of Care Determination | 0 | 0 | |||||
| Fraud/Waste & Abuse | 1 | 0 | fnol-fraud: Vendor model scoring first notice of loss for fraud indicators; high scores route to SIU and can hold payment. Observed 120 decisions in the period, 6 adverse to the consumer. direct impact: True financial: moderate | 1 | |||
| Investment/Capital Management | 0 | 0 | |||||
| Reserves/Valuations | 0 | 0 | |||||
| Catastrophe Triage | 0 | 0 | |||||
| Reinsurance | 0 | 0 | |||||
| Other Insurance Practices | 0 | 0 |
Model inventory (the six columns regulators will look for)
| Name | Description and broader system | Program area | Inherent risk | Direct consumer impact | Financial impact | Declared / observed versions | Decisions (adverse) |
|---|---|---|---|---|---|---|---|
| Underwriting Triage (uw-triage) | Scores new auto and home applications for eligibility and refers borderline risks to an underwriter. | Underwriting/Eligibility | high | True | material | 2.4 / 2.4 | 150 (26) |
| Renewal Price Optimizer (price-opt) | Recommends renewal premium within filed rating plan bounds using telematics, credit-based insurance score and territory. | Ratemaking/Rate Classification/Schedule Rating/Premium Audits | high | True | material | 7.1 / 7.2 | 168 (45) |
| Sentinel FNOL Fraud Score (fnol-fraud) | Vendor model scoring first notice of loss for fraud indicators; high scores route to SIU and can hold payment. | Fraud/Waste & Abuse | high | True | moderate | 3.1 / 3.1 | 120 (6) |
| Service Desk Assistant (agent-assist) | Generative AI assistant drafting service responses for human agents; no decision authority. | Customer Service | low | False | immaterial | 1.2 / 1.2 | 30 (0) |
Reconciliation: declared inventory versus observed decision stream
Observed but not declared: claims-fastpath v1.0 (claims, 109 decisions, 10 adverse, first seen 2026-07-02)
Version drift: price-opt declared 7.1, observed 7.2