NIST AI RMF Profile¶
This page documents how the framework maps to the four core functions of the NIST AI Risk Management Framework (AI RMF 1.0): GOVERN, MAP, MEASURE, and MANAGE.
GOVERN¶
Establishes accountability, policies, and organizational roles for AI risk.
| Framework Component | AI RMF Control |
|---|---|
GovernanceLineageAgent |
Maintains an immutable, Ed25519-signed audit log of every agent decision |
OPAClient + Rego policies |
Enforces write-gate policies before any irreversible action |
HITLSupervisorAgent |
Routes low-confidence decisions to human operators |
SecretsVault |
Manages credentials with least-privilege access |
MAP¶
Identifies and categorizes AI risks in context.
| Framework Component | AI RMF Control |
|---|---|
PurdueZoneEnforcer |
Maps each agent action to its Purdue zone impact |
UNSContextBrokerAgent |
Resolves UNS paths to physical assets, scoping potential blast radius |
AnomalyRootCauseAgent |
FMEA traversal maps anomalies to known failure modes and criticality |
RoutingPolicy |
Intent classification determines which agents handle each request |
MEASURE¶
Analyzes and assesses AI risks and impacts.
| Framework Component | AI RMF Control |
|---|---|
IABENCH-v1 |
Seven-task benchmark suite measuring F1, block-rate, hallucination, lineage completeness |
IndustrialMetrics (Prometheus) |
Real-time counters for agent requests, errors, latency, escalations |
AgentDecision.confidence |
Every decision carries a calibrated confidence score |
SafetyGuardrailAgent |
OPA pre-check + LLM evaluation quantifies risk level before action |
MANAGE¶
Prioritizes and addresses AI risks.
| Framework Component | AI RMF Control |
|---|---|
EscalationRouter |
Halts irreversible or low-confidence decisions pending human review |
WorkOrderMESAgent |
Dry-run diff preview before any CMMS/MES write; idempotency keys prevent duplication |
GovernanceLineageAgent |
NIST function tags on every OpenLineage event (GOVERN, MAP, MEASURE, MANAGE) |
PIIRedactor |
Scrubs personal identifiers from agent messages before logging |
NIST AI RMF Function Tags¶
The GovernanceLineageAgent tags every emitted OpenLineage event with its AI RMF function:
_NIST_FUNCTION_MAP = {
"read": "MEASURE",
"detect_anomaly": "MEASURE",
"evaluate_action": "MANAGE",
"block": "GOVERN",
"allow": "MANAGE",
"create_work_order":"MANAGE",
"sign": "GOVERN",
"emit_lineage": "GOVERN",
}
These tags appear in the industrialGovernance facet of each OpenLineage event and are
queryable via the governance export CLI command: