SAMPLE DOCUMENT — This is a demonstration of HAIEC's Colorado AI Act Risk Management Policy template. All data is fictional.

AI Risk Management Policy

Colorado AI Act (SB24-205) — CRS §6-1-1702(2) & §6-1-1703(2)

Organization: Meridian Financial Services
AI System: TalentFlow AI v3.2
Effective Date: January 15, 2026
Next Review: July 15, 2026
Framework: NIST AI RMF 1.0 + ISO/IEC 42001
Approved By: James Park, CTO

1. Policy Purpose & Scope

This Risk Management Policy establishes the governance framework for Meridian Financial Services' use of high-risk artificial intelligence systems as defined by the Colorado AI Act (SB24-205). This policy applies to all AI systems that make or substantially assist in consequential decisions affecting individuals in the categories defined by CRS §6-1-1701.

Risk Framework
NIST AI Risk Management Framework (AI RMF 1.0) with supplementary controls from ISO/IEC 42001:2023. This dual-framework approach satisfies the safe harbor provision under CRS §6-1-1706(3).

2. Risk Categories & Assessment

Risk Category Likelihood Impact Owner Review
Algorithmic Discrimination
Disparate impact on protected classes in hiring decisions
Medium High VP AI Governance Quarterly
Data Quality Degradation
Training data becoming stale or unrepresentative over time
Medium Medium Data Engineering Lead Monthly
Model Drift
Performance degradation as candidate population changes
High Medium ML Engineering Lead Monthly
Transparency Failure
Inadequate consumer notice or disclosure
Low High Legal & Compliance Quarterly
Human Override Failure
Recruiters over-relying on AI scores without independent judgment
Medium Medium Director of Recruiting Quarterly

3. Bias Testing Plan

Testing Methodology

Method: Disparate impact analysis using the 4/5ths rule (EEOC Uniform Guidelines), supplemented by Chi-Square and Fisher's Exact tests for statistical significance.

Frequency: Quarterly (January, April, July, October) with ad-hoc testing triggered by model updates or complaint patterns.

Protected Classes Tested: Race/ethnicity, gender, age (40+), disability status, veteran status, national origin.

Threshold: Selection rate ratio must be ≥ 0.80 for all protected classes. Any ratio below 0.80 triggers mandatory remediation review within 10 business days.

Remediation Process: If bias is detected: (1) Immediate notification to VP AI Governance, (2) Root cause analysis within 5 business days, (3) Model retraining or feature adjustment within 15 business days, (4) Re-testing to confirm remediation, (5) Documentation of findings and corrective actions in evidence bundle.

4. Discrimination Monitoring

Continuous Monitoring

Status: Enabled — real-time monitoring active since January 2026.

Metrics Tracked: Selection rate ratios by protected class, score distribution variance, false negative rates by demographic group, appeal rates by demographic group, opt-out request rates.

Alert Thresholds: Automated alerts triggered when any selection rate ratio drops below 0.85 (warning) or 0.80 (critical). Critical alerts require response within 24 hours.

Response Process: Warning alerts reviewed by AI Governance team within 48 hours. Critical alerts escalate to VP AI Governance and Legal immediately. All alerts logged in compliance evidence system.

5. Human Review & Oversight

Human-in-the-Loop Requirements

Review Level: All AI-generated candidate rankings are reviewed by a human recruiter before any candidate communication or decision.

Override Triggers: Human review is mandatory when: (1) AI score is below 40/100 and candidate has 5+ years relevant experience, (2) Candidate requests human-only review, (3) Position has fewer than 10 applicants, (4) AI system flags low confidence score.

Escalation Process: Recruiter → Hiring Manager → Director of Recruiting → VP People Operations. Each level has 2 business days to review.

Override Authority: Any hiring manager can override AI recommendations with documented justification. All overrides are logged and reviewed quarterly for patterns.

6. Incident Response

Discrimination Incident Response Protocol

Reporting Process: Any employee who suspects algorithmic discrimination must report to ai-compliance@meridianfs.com within 24 hours. Anonymous reporting available via compliance hotline (800-555-0199).

Response Timeline: Initial assessment within 48 hours. Full investigation within 15 business days. Remediation plan within 30 business days.

AG Notification: If algorithmic discrimination is confirmed, notification to the Colorado Attorney General within 90 days per CRS §6-1-1702(5) / §6-1-1703(7). Notification includes: description of discrimination, affected population estimate, remediation steps taken, and timeline for resolution.

Remediation Steps:

  • Immediately suspend AI system for affected decision category
  • Conduct root cause analysis with ML engineering and legal teams
  • Notify affected individuals within 30 days with explanation and appeal rights
  • Implement corrective measures and re-test before resuming AI-assisted decisions
  • Update risk management policy and impact assessment to reflect lessons learned
  • Conduct post-incident review at next quarterly governance meeting

7. Governance Structure

The following roles are responsible for AI risk management at Meridian Financial Services:

Role Responsibility Reports To
VP AI Governance Overall AI risk management program ownership, AG notification authority, policy approval CTO
AI Ethics Committee Quarterly review of bias testing results, policy updates, incident review Board of Directors
ML Engineering Lead Model performance monitoring, bias testing execution, remediation implementation VP AI Governance
Legal & Compliance Regulatory monitoring, consumer notice compliance, AG communication General Counsel
Director of Recruiting Human oversight enforcement, recruiter training, override review VP People Operations

Policy Approval

This AI Risk Management Policy has been reviewed and approved in accordance with CRS §6-1-1702(2) and §6-1-1703(2). It will be reviewed and updated at least semi-annually or upon material changes to AI systems.

James Park
Approved By
Chief Technology Officer
Title
1/15/2026
Date