Back to the AI Control Library
13 controls | 13 core | 0 supplemental
Why this family matters
An AI system can perform well on average while consistently disadvantaging particular people. This family addresses how organizations identify, measure and respond to differences in treatment and impact. It brings fairness into data selection, evaluation, oversight and ongoing operations rather than treating it as a one-time test before launch.
What this family covers
The controls cover bias policies, demographic and societal impact assessments, training-data relevance and provenance, fairness metrics, remediation and decision audits. They also address monitoring for changes in fairness after deployment.
How the four control types apply
Operational assessment and remediation processes make fairness concerns actionable. Technical data checks and measurement tools support that work. Governance controls establish applicable external rights and obligations. Contextual constraints include the populations, institutional setting and decision environment in which an outcome gains meaning.
Where to begin
Identify who could be affected, define relevant fairness measures, and establish how findings will change deployment decisions.
Controls in this family
All 13 controls in this family are designated core. Each page provides implementation guidance, verification criteria, evidence, escalation and a worked example. Applicability and the depth of implementation depend on the decision context and its potential effects.
Direction and external obligations
- CF02-001: Bias & Fairness Assessment Policy
- CF02-002: AI Ethics Oversight Committee
- CF02-009: Non-Discrimination & Equal Treatment Policy
- CF02-011: High-Risk AI Fairness Requirements
Impact and data foundations
- CF02-003: Demographic Impact Assessment
- CF02-004: Training Data Differentiation & Relevance
- CF02-005: Training Data Integrity for Bias Prevention
- CF02-006: Training Data Provenance & Transparency
- CF02-013: Societal Impact Assessment
Measurement, correction and accountability
- CF02-007: Bias Metrics Definition & Monitoring
- CF02-008: Bias Remediation Process
- CF02-010: Algorithmic Accountability & Decision Audit
- CF02-012: Equity & Fairness Drift Monitoring
Applying this family
Begin with the affected decision and people, then examine the workflow, data and evidence needed to understand its effects. Choose fairness measures for stated reasons, preserve uncertainty and let findings influence deployment. Review real outcomes after release and provide a meaningful route to challenge consequential decisions.
CF02-007 defines the fairness measures and their interpretation. CF02-012 follows those outcomes over time. CF09-008 addresses the operational monitoring capability that supports detection and response. These related controls serve distinct purposes and should be assessed together where applicable.
About this guidance
The requirements are the author’s synthesis for this library. They do not constitute a claim that a particular metric proves legal compliance or that all groups must have identical outcomes in every context. Qualified reviewers should determine applicable external duties.
Further reading
NIST SP 1270: Towards a Standard for Identifying and Managing Bias in Artificial Intelligence provides broader background on identifying and managing bias. It is a reference for further reading, not a validated control crosswalk.
Version 1.0