NIST AI RMF on agentic systems.
Govern, Map, Measure, Manage — applied to agents, tools, and retrieval, with a sampleable record on every material action.
How do I apply NIST AI RMF to agentic systems
Requirement, controller, artifact.
Requirement names the instrument. What GW Slate™ does is the existing controller capability. The artifact is what an auditor samples. Counsel decides what an obligation requires of a given deployment.
| Requirement | What GW Slate™ does | The artifact an auditor samples |
|---|---|---|
| Govern — identity, policy, and human authority over agentic runs. | GW Slate™ is the Sovereign Orchestrator: brand-aligned intelligence rails that hold identity, policy, runtime, and receipts over the stack you already run. Hybrid is the default posture. Quiet governance, visible proof — every material action is accountable before it becomes business truth. | The policy pack version bound to the run, and the identity recorded on every governed action. |
| Map — inventory models, agents, tools, retrieval sources, and where value belongs. | Manta Graph™ is the governed knowledge-graph layer. It maps the work, entities, context, and where value belongs before a model or tool is selected. | Delivers three artifacts: Visual Topology, Executable Schema, Compute & Token Economics. |
| Measure — a deterministic record of what ran, under which policy, with what outcome. | Receipt Rail™ is the proof layer. Every governed action emits an Actuality Receipt — who, what, policy, and outcome. | Actuality Receipt for every action — provenance, authority, logic, and outcome. |
| Manage — treat risk on the action, then hold the gate as a hard constraint. | Sovereign Orchestrator — scopes authority per action and holds policy as a hard constraint in code. | The per-action verdict — allow, deny, or human — sealed on the Actuality Receipt. |
| NIST AI 600-1 Generative AI Profile (July 2024) — generative and agentic work under the same four functions. | GW Slate™ is the modular controller and human command surface. It routes each task to the model that fits, holds authority and policy on the controller, and runs multi-agent workflows to completion. | The routing record on the receipt: which model ran, under which policy, on which workflow. |
| NIST IR 8596 ipd Cyber AI Profile — initial public draft. Treat this instrument as a draft. | An Actuality Receipt is a tamper-evident cryptographic record emitted by Receipt Rail™ proving an AI action's provenance, authority, logic, and outcome — auditable proof a regulator or counterparty can check. | The tamper-evident Actuality Receipt: provenance, authority, logic, and outcome. |
Last reviewed: 2026-09-08
Questions
01How do I apply NIST AI RMF to agentic systems?
Run the four functions on the action: Manta Graph™ maps models, agents, tools, and retrieval; GW Slate™ holds identity, policy, runtime, and receipts; Receipt Rail™ emits the Actuality Receipt an auditor samples for Measure and Manage. NIST AI RMF 1.0 is voluntary.
02Does NIST AI RMF issue a certificate?
NIST AI RMF 1.0 is a voluntary four-function program — Govern, Map, Measure, Manage. The record an organization keeps is the Measure and Manage evidence the controller produces toward that program.
03What is NIST AI 600-1?
NIST AI 600-1 is the Generative AI Profile, published in July 2024. It applies the RMF functions to generative work. GW Slate™ routes each task to the model that fits and records that route on the Actuality Receipt.
04What is NIST IR 8596 ipd?
NIST IR 8596 ipd, Cybersecurity Framework Profile for Artificial Intelligence (Cyber AI Profile), is an initial public draft. The controller still produces the same Measure record: a tamper-evident Actuality Receipt for every governed action.
Continue the work
Diagnostic Sprint, then Department, then Sovereign Enterprise.
$7,500 standard / $10,000 regulated, 100% creditable.
What the certifiable AIMS requires of AI agents.
What the Act requires now, including Article 50.
Operational controls inside the workflow.
The map an auditor reads before the policy gate returns a verdict.