Colorado AI Act

The Colorado AI Act (SB21-169) regulates automated decision systems used in consequential decisions affecting Colorado residents.

Overview

Aspect	Details
Effective Date	February 1, 2024
Scope	Consequential decisions in insurance
Regulator	Colorado Division of Insurance
Penalties	Per-violation fines, license revocation

Key Requirements

§ 6-1-1303: Impact Assessments

"A deployer shall complete an impact assessment for each high-risk AI system"

Requirements:

• Annual impact assessments

• Risk identification

• Mitigation documentation

ATHENA Solution:

• Trend analysis over time

• Risk scoring

• Automated report generation

const report = await athena.export.generate({
  template: 'colorado',
  format: 'pdf',
  startDate: '2025-01-01',
  endDate: '2025-12-31'
});

§ 6-1-1304: Consumer Notification

"A deployer shall provide notice to a consumer... that an AI system was used"

ATHENA Solution:

• Decision logging with notification tracking

• Audit trail for compliance proof

§ 6-1-1306: Bias Testing

"A deployer shall test for unfair discrimination"

ATHENA Solution:

• Continuous bias detection

• Subgroup performance analysis

• Disparity alerts

Impact Assessment Contents

1. System Description

• Purpose of AI system

• Decision types made

• Data inputs used

2. Risk Analysis

• Identified risks to consumers

• Potential for discrimination

• Harm severity assessment

3. Performance Metrics

• Accuracy by subgroup

• Bias detection results

• Trust calibration scores

4. Mitigation Measures

• Controls implemented

• Monitoring procedures

• Remediation actions

5. Trend Analysis

• Year-over-year comparison

• Improvement tracking

• Persistent issues

Implementation

Step 1: Log All Decisions

await athena.audit.log({
  userId: 'agent_123',
  aiRecommendation: 'Deny coverage',
  userDecision: 'Deny coverage',
  outcome: 'claim_denied',
  context: 'Auto insurance application',
  metadata: {
    state: 'CO',
    product: 'auto_insurance'
  }
});

Step 2: Track Subgroups

const result = await athena.bias.detect({
  decisionId: 'dec_123',
  subgroups: {
    ageGroup: '25-34',
    gender: 'male',
    zipCode: '80202'  // Colorado
  }
});

Step 3: Monitor Trends

const trend = await athena.trustScore.trend({
  userId: 'agent_123',
  days: 365
});

console.log('Year-over-year trend:', trend.dataPoints);

Step 4: Generate Annual Report

const impact = await athena.export.generate({
  template: 'colorado',
  format: 'pdf',
  startDate: '2025-01-01',
  endDate: '2025-12-31',
  filters: {
    metadata: { state: 'CO' }
  }
});

Annual Assessment Schedule

Year	Due Date
2024	January 31, 2025
2025	January 31, 2026
2026	January 31, 2027

Consumer Rights

Colorado residents have the right to:

Right	ATHENA Support
Know AI was used	Decision logging
Explanation of decision	Audit trail with context
Appeal adverse decisions	Decision history
Opt-out (where applicable)	Tracking metadata

Sample Impact Assessment

COLORADO AI ACT IMPACT ASSESSMENT
Organization: Mountain Insurance Corp.
Assessment Period: 2025

SYSTEM DESCRIPTION
==================
System Name: Underwriting Risk Scorer
Purpose: Auto insurance risk assessment
Decisions Made: 45,678
Colorado Residents Affected: 12,456

RISK ANALYSIS
=============
Identified Risks:
1. Age discrimination (65+ applicants)
   - Severity: MEDIUM
   - Likelihood: LOW
   - Mitigation: Age-adjusted models

2. Geographic bias (rural vs urban)
   - Severity: LOW
   - Likelihood: MEDIUM
   - Mitigation: Zip code normalization

PERFORMANCE METRICS
==================
Overall Accuracy: 87.3%
Subgroup Analysis:
  - age_group:18-24: 84.2%
  - age_group:25-34: 88.1%
  - age_group:35-44: 89.0%
  - age_group:45-54: 88.5%
  - age_group:55-64: 86.7%
  - age_group:65+: 82.1% ⚠️ Below threshold

Bias Detected: 1 subgroup (age_group:65+)
Disparity: 5.2 percentage points

TREND ANALYSIS
==============
Year-over-Year Comparison:
  - 2024: 1 subgroup with disparity
  - 2025: 1 subgroup with disparity (same group)
  - Status: NO IMPROVEMENT

REMEDIATION PLAN
===============
1. Retrain model with balanced age representation
2. Implement age-adjusted accuracy thresholds
3. Quarterly monitoring of 65+ subgroup
4. Target: Reduce disparity to <3% by Q2 2026

COMPLIANCE STATUS: COMPLIANT WITH FINDINGS

---

Next: FDA AI/ML Guidelines