BB-UW-0091 FLAG
On 2026-10-23, uw-triage v2.4 (an internally developed system) produced 'decline' in a TX auto underwriting transaction. Inputs included application data, claims history, consumer report, credit, geography. A consumer report was used. A human reviewed the output by a underwriter. The outcome was adverse to the consumer; a notice was sent on 2026-10-28 with specific reasons. RegBench applied 11 rule(s) in force on 2026-10-23 and recorded 1 finding(s); the most serious is 'Geography and credit used together in an adverse decision (proxy pattern to test at portfolio level)' (medium, L6).
FLAG: 1 finding(s) under rules 2026.10+dbe5ddf6, corpus 2026-10-30-5f07c521. Routed to the carrier's compliance reviewer with cited findings.
Findings (1)
Geography and credit used together in an adverse decision (proxy pattern to test at portfolio level)
Finding. The decision used both geographic and credit inputs and was adverse to the consumer; the interaction of these inputs is a recognised proxy pattern for protected characteristics.
Basis. Doctrine: the bulletin's concern with unfair discrimination extends to proxies, and the Supplement's Exhibit C item 10a asks whether inputs, derived variables and outputs are evaluated "for potential proxy relationships"; a geography-by-credit interaction should be covered by portfolio-level outcome testing and the result referenced from the case file.
Required action. Confirm the system's portfolio outcome testing covers the geography-credit interaction; link the result.
Excerpt: naic-ai-bulletin-s3-1-ais-program · NAIC model guidance; binding in a state only as adopted (see adoptions register) · in force from 2023-12-04
"all Insurers authorized to do business in this state are expected to develop, implement, and maintain a written program (an 'AIS Program') for the responsible use of AI Systems that make, or support decisions related to regulated insurance practices. The AIS Program should be designed to mitigate the risk of Adverse Consumer Outcomes ... The Department also encourages the development and use of verification and testing methods to identify errors and bias in Predictive Models and AI Systems, as well as the potential for unfair discrimination in the decisions and outcomes resulting from the use of Predictive Models and AI Systems."
Point in time
- Law as of
- 2026-10-23
- Rules version
- 2026.10+dbe5ddf6
- Corpus snapshot
- 2026-10-30-5f07c521
- Engine
- regbench-engine/0.4.0 · deterministic-1
- Fingerprint
- 2e721a05917cbd7825c2af94
- Evaluated
- 2026-10-30 11:23:22 UTC
- Exhibit C case file
- yes
Ledger
- Entry
- #91 · assessment · recorded 2026-10-23T19:31:42+00:00
- Hash
- 6bbecd1e12300ff8a89decfa92af2b1d4b823cb62874fed65bbff494003507bc
- Previous
- 2a2d9076d74636ee742243252f24775dbce40cf65d0808a885bf9210d1cc3639
- Stored copy intact
- yes, hashes match
Carrier review
Resolved by underwriter · outcome remediated · 58.8 hours after it opened · Resolved in the carrier's review queue (synthetic).
Recorded facts
- AI system
- uw-triage v2.4 · internal · high-risk class: underwriting
- Governance attested
- True
- Testing report
- test://bluebonnet/uw-triage/2026Q3
- Model documentation
- doc://bluebonnet/uw-triage/2.4
- Feature classes
- application_data, claims_history, consumer_report, credit, geography
- Protected factors used
- none
- External data (ECDIS)
- False
- Consumer report / credit
- True / True
- Data sources
- cra://demo-bureau/report
- Decision
- decline · score 0.964 · codes R12, R31
- Human review
- yes (underwriter)
- Adverse to consumer
- True
- Notice
- sent 2026-10-28 · reasons: True · specific: True · external data disclosed: False
- Appeal channel
- –
- Consumer token
- tok_a605611673