AI Governance

EU AI Act Article 13 conformance

ARIA publishes a transparency statement with every high-risk decision path: model identity, confidence score, ethics-gate threshold applied, frameworks invoked, and upstream citations. The full per-decision log is exportable from the admin surface (see Model Transparency).

Ethics-gate thresholds

Per-tenant defaults live in orgs.ethics_gate_config. Operators can tighten but never loosen below the regulation floor.

Decision-pattern catalogue

ARIA selects from a curated catalogue of evidence-based decision frameworks on every decision (selection rationale and citations available under Trust → AI Governance docs). Every selection is logged to the decision trajectory.

Sample explainability record

{
  "decision": "route_to_agent",
  "agent": "compliance-advisor",
  "confidence": 0.93,
  "decision_patterns_applied": [
    "behavior-change-pattern",
    "friction-reduction-pattern",
    "gdpr-article-13"
  ],
  "ethics_gate": {
    "threshold": 0.92,
    "passed": true,
    "reason": "tenant_regulation=gdpr"
  },
  "trigger_stack": [
    "social-proof-customer-count",
    "retention-risk-signal"
  ]
}

Brand-voice extraction (ADR-058)

ARIA's brand-voice extractor removes AI-tells on every customer-facing output. Example: