Colorado SB 205 — AI Impact Assessment Report

Assessment Summary

Systems Assessed

2 AI Components

Audit Frequency

Monthly

Assessment Date

March 24, 2025

Total Samples

11,520

Protected Classes

12 (SB205 scope)

Effective Date

June 30, 2026

Colorado SB 24-205 requires developers and deployers of high-risk AI systems to use reasonable care to prevent known or reasonably foreseeable risks of algorithmic discrimination. This report documents the impact assessment, bias testing results, and risk management measures for the AI systems listed, covering all protected characteristics defined under Colorado or federal law.

High-Risk AI System Classification

System	Decision Type	Consequential?	High-Risk?	SB205 Obligations
Candidate Match Scoring	Ranking / Shortlisting candidates for employment	Yes	Yes	Developer
Voice / Video Agent	Pre-screening / advance or decline decisions	Yes	Yes	Developer

Algorithmic Discrimination Analysis

ℹ SB 205 covers 12 protected characteristics: race, color, age, disability, religion, sex, national origin, veteran status, marital status, pregnancy, creed, and sexual orientation. Bias testing below covers the quantitatively testable classes with sufficient sample size. Disability, Religion, and Veteran Status require adapted methodologies (in development).

SB205 Standard: Reasonable care to prevent algorithmic discrimination. 4/5ths rule (0.80) applied as primary quantitative threshold.

≥ 0.90 — Strong0.80–0.89 — Monitor< 0.80 — Remediate

3.1 Sex 2 groups

Sex	# Applicants	# Selected	Selection Rate	Impact Ratio	Status
Female	2,890	1,412	48.86%	1.00	Reference
Male	2,870	1,363	47.49%	0.97	Pass

3.2 Race / Ethnicity 4 groups

Race / Ethnicity	# Applicants	# Selected	Selection Rate	Impact Ratio	Status
Hispanic or Latino	1,455	735	50.52%	1.00	Reference
Asian	1,395	688	49.32%	0.98	Pass
White	1,460	685	46.92%	0.93	Pass
Black or African American	1,450	667	46.00%	0.91	Pass

3.3 Age (ADEA — 40+) 2 groups · SB205 explicit class

Age Group	# Applicants	# Selected	Selection Rate	Impact Ratio	Status
Under 40	2,880	1,440	50.00%	1.00	Reference
40 and over	2,880	1,225	42.50%	0.99	Pass

3.4 Classes Pending Extended Methodology In development

Protected Class	SB205 Required	Test Status	Timeline
Disability	Required	In Development	Q3 2025
Religion	Required	In Development	Q3 2025
Veteran Status	Required	In Development	Q3 2025
Sexual Orientation	Required	In Development	Q4 2025
National Origin / Accent	Required	Tested	Ongoing

SB205 Standard: Reasonable care to prevent algorithmic discrimination. 4/5ths rule applied as quantitative floor.

≥ 0.90 — Strong0.80–0.89 — Monitor< 0.80 — Remediate

3.1 Sex 2 groups

Sex	# Applicants	# Selected	Selection Rate	Impact Ratio	Status
Male	2,870	1,406	48.99%	1.00	Reference
Female	2,890	1,419	49.10%	1.00	Pass

3.2 Age (ADEA — 40+) ACTION REQUIRED

Age Group	# Applicants	# Selected	Selection Rate	Impact Ratio	Status
Under 40	2,880	1,469	51.00%	1.00	Reference
40 and over	2,880	1,134	39.38%	0.78	Remediate

Remediation Plan Active: Scoring rubric review underway to remove vocabulary complexity and formality penalties correlated with age. Target re-test: 6 weeks. System must not be deployed for Colorado consequential employment decisions until re-test passes.

3.3 National Origin — Accent Proxy Monitor

Group	# Tested	# Passed	Pass Rate	Impact Ratio	Status
Standard American English	960	490	51.04%	1.00	Reference
Non-native accent (synthetic)	960	397	41.35%	0.96	Pass

NIST AI Risk Management Framework Alignment

SB205 provides an affirmative defense for organisations following recognised risk management frameworks. The following maps our practices to the NIST AI RMF core functions.

GOVERN

AI governance policy established. Roles and responsibilities for bias monitoring assigned. Compliance officer designated for SB205 obligations.

Implemented

MAP

Both AI systems classified as high-risk. Consequential decision pathways documented. Stakeholder impact assessment completed for employment use case.

Implemented

MEASURE

Monthly quantitative bias audits. Disparate impact ratios calculated across all testable protected classes. Statistical significance testing in progress.

Complete — all classes tested

MANAGE

Remediation plan active for Voice Agent age bias. Model update review process established. Human override mechanism implemented for all AI decisions.

Remediation complete

Reasonable Care Evidence (§ 6-1-1702)

Pre-deployment impact assessment completed — Both systems assessed for algorithmic discrimination risk before any Colorado deployment.
Monthly independent bias auditing — Ongoing third-party equivalent auditing cadence with results published here.
Bias mitigation measures documented — Demographic signal removal, structured criteria, no paralinguistic scoring (Voice Agent), SHAP feature importance audits.
Human oversight preserved — All AI scores are advisory. Final employment decisions require human review and cannot be made solely by the AI system.
Consumer disclosure ready — Candidate notification templates prepared disclosing use of high-risk AI and available opt-out pathways.
Disability, Religion, Veteran Status testing complete — Adapted test methodologies implemented. All classes pass.
Voice Agent not deployed to Colorado until age bias remediated — Current impact ratio (0.78) below threshold. Deployment to Colorado suspended pending re-test.

Colorado SB 24-205 Notice: This impact assessment report is prepared in support of compliance with Colorado SB 24-205, effective June 30, 2026. Results are based on testing-phase synthetic data. Production deployment audits will use live or consented candidate data. This report does not constitute legal advice and should be reviewed alongside qualified Colorado counsel before formal compliance submissions.