# Algorithmic impact assessment
> Source: https://legalcode.md/skills/legalcode-algorithmic-impact-assessment | [HTML version](https://legalcode.md/skills/legalcode-algorithmic-impact-assessment)

Conduct comprehensive algorithmic impact assessments (AIAs) for AI systems that make or substantially influence consequential decisions affecting individuals.

Conduct comprehensive algorithmic impact assessments (AIAs) for AI systems that make or substantially influence consequential decisions affecting individuals. Covers system classification and purpose documentation, training data and data governance analysis, fairness metrics evaluation (demographic parity / impact ratio, equalized odds, calibration / predictive parity, individual fairness, counterfactual fairness), disparate impact testing across all legally relevant protected classes, proxy feature and protected characteristic inference analysis, bias detection and root cause analysis, transparency and explainability audit (model card completeness, LIME/SHAP/feature importance), human oversight mechanism assessment, appeal and contestation procedure review, and ongoing monitoring framework design. Aligned with EU AI Act (Regulation 2024/1689) Articles 9, 10, 14, 27, and Annex III high-risk system categories; NYC Local Law 144 bias audit requirements for automated employment decision tools; Colorado AI Act (SB 24-205, effective June 30 2026) consequential decision impact assessment obligations; California CPRA Automated Decisionmaking Technology (ADMT) Regulations (effective January 1 2026); NIST AI Risk Management Framework (AI RMF 1.0) MAP and MEASURE functions; IEEE 7003-2024 algorithmic bias considerations; and ISO/IEC TR 24027 bias identification guidance. Integrates EEOC Title VII adverse impact guidance, CFPB ECOA adverse action and less-discriminatory-alternative requirements, FTC Section 5 unfairness doctrine, and Fair Housing Act / Fair Credit Reporting Act overlays. Use when deploying AI systems subject to regulatory transparency or accountability requirements; preparing for EU AI Act Article 27 Fundamental Rights Impact Assessment (FRIA); conducting NYC LL 144 annual bias audits; satisfying Colorado SB 24-205 pre-deployment impact assessment obligations; completing California CPRA ADMT risk assessments; assessing employment, credit, housing, healthcare, insurance, education, or criminal justice automated decision systems for disparate impact; or building an organizational algorithmic impact assessment program. Produces comprehensive AIA reports with fairness metrics scorecards, disparate impact findings, bias root cause analysis, transparency audit, human oversight evaluation, remediation roadmaps, compliance mapping, and Glass Box audit trails suitable for regulatory submission and internal governance.

This skill conducts systematic algorithmic impact assessments of AI systems that make or substantially influence decisions with material legal or practical consequences for individuals and groups. It identifies, quantifies, and supports remediation of risks of unfair, discriminatory, opaque, or inadequately overseen automated decision-making. **Covers:** - AI system type classification and consequential decision category determination - Applicable regulatory framework identification (EU AI Act Annex III, NYC LL 144, Colorado SB 24-205, California CPRA ADMT, NIST AI RMF, EEOC, CFPB, FHA, FTC) - Training data governance analysis (representativeness, completeness, bias examination) - Fairness metrics analysis across ten assessment dimensions: 1. Demographic parity and group-level impact ratios 2. Equalized odds and error rate parity 3. Calibration and predictive parity 4. Individual fairness and decision consistency 5. Proxy feature and protected characteristic inference 6. Transparency and model card completeness 7. Human oversight and appeal procedure adequacy 8. Ongoing monitoring and feedback loop prevention 9. Sector-specific and vulnerable group protections (conditional) 10. Disparate impact and intersectional discrimination (conditional) - Protected class coverage mapping (US federal, NYC, Colorado, California, EU Charter, GDPR Art. 9) - Bias root cause analysis (training data, feature engineering, model architecture, deployment context, feedback loops) - Transparency and explainability audit (LIME, SHAP, ...

## Jurisdictions

Global



---
*Legalcode -- legalcode.md*