← All specifications

NIST AI Risk Management Framework — ponens Policy Pack

This pack maps the NIST AI Risk Management Framework (AI RMF 1.0) — the foundational, voluntary US framework for managing AI risk — onto computable ponens policies. It is squarely on-mission for ponens (AI governance) and brings a shape distinct from the financial-regulator packs: a lifecycle risk-management framework rather than a runtime or conduct regime.

Source: NIST AI 100-1, AI Risk Management Framework 1.0 (26 Jan 2023). https://www.nist.gov/itl/ai-risk-management-framework

Why this maps onto ponens

The AI RMF is organised around four functions, which become the pack’s groups, and seven trustworthiness characteristics, which land under MEASURE:

Trustworthiness characteristics (under MEASURE): valid & reliable · safe · secure & resilient · accountable & transparent · explainable & interpretable · privacy-enhanced · fair (harmful bias managed).

ponens turns the subcategories into policies over an AI system’s lifecycle record — making voluntary AI-RMF adoption auditable rather than aspirational.

AI RMFponens
The AI system’s lifecycle / governance recordthe trace
A function’s subcategorya policy (temporal formula)
Core requirement vs supporting practiceerror (Red) / warning (Amber)

Trace model

Reuses existing vocabulary — Deploy, Decision, Output, Incident, Plan, EditFile (no new action types). Per-action predicates include risk_policy_in_place, accountable_role, legal_requirements_mapped, context_established, system_categorized, impacts_characterized, validity_reliability_measured, safety_measured, security_resilience_measured, privacy_measured, transparency_measured, explainability_measured, fairness_measured, bias_managed, tracked, risks_prioritized, identified_risk, risk_treated, responded, recovered, third_party_ai, thirdparty_risk_managed.

Worked traces: examples/nist_ai_rmf/governed.json (15/15 Green) and violating.json (7 Red + 4 Amber — GOVERN satisfied, MAP/MEASURE/MANAGE largely missing). Run ponens trace check <file>.

The pack

errorRed; warningAmber.

GOVERN (conformance)

PolicyFormulaRAG
rmf_govern_policies_in_placeG((Deploy ∨ Decision) → P(risk_policy_in_place))R
rmf_govern_accountability_assignedG((Deploy ∨ Decision) → accountable_role)R
rmf_govern_legal_requirements_mappedG(Deploy → P(legal_requirements_mapped))A

MAP (conformance)

PolicyFormulaRAG
rmf_map_context_establishedG(Deploy → P(context_established))R
rmf_map_system_categorizedG(Deploy → P(system_categorized))R
rmf_map_impacts_characterizedG(Deploy → P(impacts_characterized))R

MEASURE — Trustworthiness (safety / security / auditability)

PolicyFormulaRAG
rmf_measure_validity_safetyG(Deploy → P(validity_reliability_measured ∧ safety_measured))R
rmf_measure_security_privacyG(Deploy → P(security_resilience_measured ∧ privacy_measured))R
rmf_measure_transparency_explainabilityG(Deploy → P(transparency_measured ∧ explainability_measured))A
rmf_measure_fairness_biasG(Deploy → P(fairness_measured ∧ bias_managed))R
rmf_measure_ongoing_trackingG(Output → tracked)A

MANAGE (workflow / auditability)

PolicyFormulaRAG
rmf_manage_risks_prioritizedG(Deploy → P(risks_prioritized))R
rmf_manage_risk_treatedG(identified_risk → P(risk_treated))R
rmf_manage_incident_responseG(Incident → F(responded ∧ recovered))R
rmf_manage_thirdparty_risksG(third_party_ai → P(thirdparty_risk_managed))A

Aggregation

ponens trace check aggregates the pack: any error fail ⇒ Red (no risk policies/accountability, unmapped impacts, unmeasured safety/fairness, an unhandled incident); else any warning fail ⇒ Amber (legal mapping, transparency/explainability measurement, ongoing tracking, third-party risk); else Green.

Notes

The AI RMF Playbook provides finer-grained subcategory actions, and the Generative AI Profile (NIST AI 600-1) adds GenAI-specific risks; both could extend this pack as further policies. This pack covers the AI RMF 1.0 core. It pairs naturally with the agentic-runtime (FIX) and supervisory (IOSCO/ESMA) packs — GOVERN/MAP/MEASURE/MANAGE is the lifecycle frame within which those operate.