AI fairness control library32
HighGOV-01
Fairness policy and decision scope
Define applicable insurance decisions, consumer populations, harms, legal standards, risk appetite, and approval authority.
HighGOV-02
Consumer impact assessment
Assess how the AI-supported process can affect access, price, coverage, claim outcomes, service, privacy, dignity, and recourse.
CriticalGOV-03
Accountability and independent challenge
Assign business, actuarial, data, compliance, legal, operations, and technical ownership with qualified independent review.
CriticalGOV-04
Fairness acceptance and stop conditions
Document thresholds, exceptions, residual risk, expiration, monitoring, escalation, and conditions that pause or limit use.
HighDAT-01
Population and coverage analysis
Compare development, validation, production, and affected populations by relevant segments, geography, product, channel, and time.
HighDAT-02
Data provenance and permitted use
Record source, collection context, consent or authority, transformations, lineage, quality, retention, and approved purpose.
CriticalDAT-03
Label and outcome quality
Test whether labels and outcome definitions are accurate, timely, consistently applied, and appropriate for the insurance decision.
CriticalDAT-04
Missingness and measurement error
Measure missing, delayed, stale, imputed, and systematically mismeasured data by segment and operating context.
HighMET-01
Segment and intersection selection
Define legally and operationally relevant groups, intersectional views, sample limits, privacy controls, and testing cadence.
HighMET-02
Outcome and selection-rate testing
Compare favorable and adverse outcomes using denominators that match the decision, eligibility population, and period.
CriticalMET-03
Error and service-quality testing
Compare false positives, false negatives, error severity, latency, rework, and service quality across segments.
CriticalMET-04
Calibration and uncertainty
Assess whether scores and probabilities have comparable meaning, reliability, and uncertainty across relevant segments.
HighPRX-01
Protected attribute and proxy review
Identify direct, inferred, correlated, geographic, behavioral, and composite variables that may encode protected characteristics.
HighPRX-02
Business necessity and actuarial relevance
Document the relationship between each material variable and the insurance outcome, including limitations and alternative explanations.
CriticalPRX-03
Less discriminatory alternatives
Search for feasible variables, methods, thresholds, workflows, or human controls that meet the legitimate objective with less disparity.
CriticalPRX-04
Counterfactual and sensitivity testing
Test how outcomes change when sensitive or proxy-linked attributes and related conditions vary while relevant risk factors remain controlled.
HighUWP-01
Eligibility and coverage review
Test AI-supported eligibility, declination, nonrenewal, coverage, tier, and referral outcomes across affected populations.
HighUWP-02
Pricing and rating review
Assess indicated and charged price, discounts, surcharges, tiers, and error consequences using applicable actuarial and legal standards.
CriticalUWP-03
External consumer data review
Evaluate relevance, accuracy, provenance, recency, correction rights, and differential impact of external consumer data and scores.
CriticalUWP-04
Distribution and access review
Assess channel, digital access, language, disability, geography, device, and assistance needs that affect product access or service.
HighCLM-01
Claims triage and severity review
Test prioritization, routing, reserve support, complexity, and service outcomes without allowing protected or proxy factors to drive unfair treatment.
HighCLM-02
Fraud and investigation review
Compare alerts, investigation referrals, false positives, burdens, delays, and closures across relevant segments.
CriticalCLM-03
Settlement and communication review
Assess offer support, negotiation prompts, communications, documentation requests, and timeliness for differential outcomes or burdens.
CriticalCLM-04
Human review and automation-bias controls
Measure overrides, disagreements, review time, escalation quality, and whether staff rely on AI without adequate challenge.
HighREC-01
Decision explanation and reason fidelity
Provide reasons that accurately reflect the factors and rules that materially influenced the individual insurance outcome.
HighREC-02
Notice and accessibility
Deliver timely, understandable, accessible, language-appropriate notices through channels consumers can use.
CriticalREC-03
Correction dispute and appeal
Enable consumers and staff to correct data, dispute an outcome, obtain human review, and track resolution without retaliation or added burden.
CriticalREC-04
Complaint and harm signals
Connect complaints, appeals, legal inquiries, accommodations, cancellations, and corrected decisions to the AI system and release.
HighOPS-01
Production fairness monitoring
Monitor outcome, error, calibration, service, complaint, appeal, and override measures by relevant segments and release.
HighOPS-02
Change and drift revalidation
Repeat affected fairness tests when data, models, prompts, vendors, rules, products, populations, workflows, or laws change.
CriticalOPS-03
Vendor evidence and accountability
Require data, methods, testing, limitations, change notice, audit support, correction, incident cooperation, and exit rights.
CriticalOPS-04
Remediation and affected consumer review
Contain the issue, identify affected decisions, correct outcomes, communicate as required, retest controls, and close root causes.