Insurance AI Bias Attestation

Gated two-step chain for annual CO SB 21-169 / Reg. 10-1-1 bias attestation. Step 1 tests aggregate BIFSG regression outputs against the two statutory prongs: statistical prong (p<0.05 AND marginal effect ≥5pp) and premium prong (model-predicted premium ≥5% above group average). A §21.4 gate routes to Step 2 when bias is detected, triggering NAIC AIS program readiness assessment (6-dimension, 0-18 score) to surface corrective action items for the Dec 1 annual filing. No bias detected exits as an attestation pass. Zero PII — aggregate statistics only.

OpenChainGraph · 2 Steps compliance mandate · §21.4 Gate Hash-Anchored §4 chain_depth:2 CO SB 21-169 · NAIC AIS · Zero PII

Chain Stages · 2 Steps · 1 Decision Gate (§21.4)

1 ROOT & GATE node §21.4 gate

CO SB 21-169 BIFSG Bias Threshold Tester art-239-test-bifsg-bias-thresholds

Tests aggregate BIFSG regression outputs against CO SB 21-169 / Reg. 10-1-1 bias detection thresholds. Statistical prong: p<0.05 AND |marginal effect| ≥5pp. Premium prong: model-predicted premium ≥5% above group-weighted average. Annual attestation deadline: Dec 1. Gate: bias_detected=true routes to NAIC AIS readiness assessment. Default (no bias) exits as attestation pass. Aggregate statistics only.

△ Gate: if /bias_detected = true → Step 2 (NAIC AIS assessment) | default (false) → END (attestation pass)

MCP Call · test_bifsg_bias_thresholds

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "test_bifsg_bias_thresholds",
    "arguments": {
      "model_purpose": "pricing",
      "line_of_business": "life_health",
      "p_value": 0.03,
      "marginal_effect_pp": 7.2,
      "predicted_premium": 1320,
      "group_avg_premium": 1200,
      "attestation_year": 2025,
      "n_observations": 5000
    }
  },
  "id": 1
}

↓

2 TERMINAL node

NAIC AIS Program Readiness Assessor art-240-assess-naic-ais-program-readiness

Scores insurance AI program against NAIC AI Model Bulletin (2020/2023) and AIS Evaluation Tool (2024). 6 dimensions: governance, risk management, data management, model documentation, human oversight, fairness testing. Score 0-18: GREEN ≥78% (14-18), YELLOW ≥44% (8-13), RED <44% (0-7). Priority gaps drive corrective action items for annual attestation filing. Terminal stage — exports composite §4 artifact.

MCP Call · assess_naic_ais_program_readiness

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "assess_naic_ais_program_readiness",
    "arguments": {
      "governance": 2,
      "risk_management": 1,
      "data_management": 2,
      "model_documentation": 2,
      "human_oversight": 3,
      "fairness_testing": 1
    }
  },
  "id": 2
}

Insurance AI Bias Attestation