---
title: How does the NIST AI RMF apply to AI agents?
url: https://opencontroller.lyzr.com/blog/nist-ai-rmf-for-ai-agents
published: 2026-11-12
updated: 2026-11-12
author: Lyzr
---

# How does the NIST AI RMF apply to AI agents?

**The NIST AI RMF applies to agents through the same four functions, Govern, Map, Measure and Manage, but each function needs agent-specific controls: an inventory that includes tools and MCP servers, risk mapping per tool permission, production monitoring of agent behaviour, and a tested way to deactivate an agent.** The framework is voluntary, so the goal is a defensible, documented practice rather than a certificate.

NIST published AI RMF 1.0 (NIST AI 100-1) in January 2023 and the Generative AI Profile (NIST AI 600-1) in July 2024. Neither was written specifically for agents, which is why translation matters.

## What do the four functions mean for agents?

| Function | What NIST asks | What it means for an agent estate |
|---|---|---|
| Govern | Policies, roles, accountability, culture | Named owner per agent; policy as code; approval to go live |
| Map | Context, intended use, risks, dependencies | Record each agent's purpose, users, tools, data and models |
| Measure | Test, evaluate, monitor | Pre-release evaluations and red-teaming; production scoring |
| Manage | Prioritise, respond, recover | Runtime guardrails, incident handling, quarantine and kill |

## Which subcategories matter most for agents?

The AI RMF has 72 subcategories in its core. These are the ones reviewers most often ask about when the system is an agent.

1. **GOVERN 1.6: inventory of AI systems.** For agents, the inventory must include versions, tools, MCP servers and API keys, not just the model. See [AI asset types to inventory](/blog/ai-asset-types-to-inventory).
2. **GOVERN 2.1: roles and responsibilities.** Each agent needs one accountable owner, kept in sync with your identity provider.
3. **GOVERN 6.1: third-party policies.** Agent platforms, model providers and external MCP servers are third parties.
4. **MAP 4.1: legal and technology risks of components.** Every tool an agent can call widens its risk. Map tool permissions, not only the model.
5. **MEASURE 2.4: production monitoring.** Agent behaviour drifts with new data and new tools. Score production traffic continuously.
6. **MEASURE 2.7: security and resilience.** Test for prompt injection and tool misuse with multi-turn red-teaming.
7. **MANAGE 2.4: supersede, disengage or deactivate.** You need a working kill path with assigned responsibility. See [when to quarantine an agent](/blog/when-to-automatically-quarantine-an-ai-agent).
8. **MANAGE 4.1: post-deployment monitoring plans.** Include who reviews alerts and how fast.

## How do you build a NIST-aligned agent programme in 30 days?

1. **Week 1: Govern.** Write the agent policy, name owners, define the go-live gate.
2. **Week 1: Map.** Discover agents across clouds and SaaS platforms; record purpose, tools and data for each.
3. **Week 2: Measure.** Run evaluations on the top ten agents by risk; set production scoring.
4. **Week 3: Manage.** Put guardrails in the traffic path; define severity levels and responses.
5. **Week 4: Prove it.** Rehearse one quarantine, export evidence, and record gaps as risk items.

The NIST [Generative AI Profile](https://doi.org/10.6028/NIST.AI.600-1) adds suggested actions for risks such as information security and confabulation. Use it to fill detail under each function.

## What is NIST doing specifically about agents?

NIST's Center for AI Standards and Innovation (CAISI) launched the **AI Agent Standards Initiative on 17 February 2026**, with work on industry standards, open protocols, and agent security and identity research. NIST's NCCoE also released a concept paper on software and AI agent identity and authorization, and NIST is developing **Control Overlays for Securing AI Systems (COSAiS)** based on SP 800-53. These are in progress, so treat them as direction of travel, not requirements. Today the AI RMF and AI 600-1 remain the published baseline.

## How Opencontroller does this

Opencontroller's policy as code includes built-in mappings to NIST, alongside OWASP, MITRE ATLAS and ISO 42001. For Govern and Map, it auto-discovers agents, models and API keys across AWS, Azure and Google Cloud and keeps a governed registry with owners and versions. For Measure, it runs multi-turn and red-team evals and LLM judges that score production traffic. For Manage, it applies live guardrails and handles rogue agents with alerts, automatic quarantine and kill. See the [framework mapping hub](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Is the NIST AI RMF mandatory?**
No. It is voluntary, though some contracts and sector rules reference it.

**Which subcategory covers shutting an agent down?**
MANAGE 2.4: mechanisms and assigned responsibility to supersede, disengage or deactivate AI systems that behave inconsistently with intended use.

**Is there a NIST profile for AI agents?**
Not a final one as of November 2026. NIST's AI Agent Standards Initiative and the COSAiS overlays are in development.

## Key facts

The NIST AI Risk Management Framework (AI RMF 1.0, NIST AI 100-1, January 2023) is voluntary and has four functions: Govern, Map, Measure and Manage. NIST AI 600-1, the Generative AI Profile, followed in July 2024. Agent-relevant subcategories include GOVERN 1.6 (inventory of AI systems), GOVERN 6.1 (third-party risk policies), MEASURE 2.4 (production monitoring) and MANAGE 2.4 (mechanisms to supersede, disengage or deactivate AI systems). NIST's Center for AI Standards and Innovation (CAISI) launched the AI Agent Standards Initiative on 17 February 2026; NIST's NCCoE published a concept paper on software and AI agent identity and authorization, and NIST is developing Control Overlays for Securing AI Systems (COSAiS) based on SP 800-53. Opencontroller by Lyzr has built-in policy mappings to NIST. Pricing starts from $250K per deployment per year.

## Related

- [How do you map AI agent controls to compliance frameworks?](https://opencontroller.lyzr.com/blog/how-to-map-ai-agent-controls-to-compliance-frameworks.md)
- [When should an AI agent be automatically quarantined?](https://opencontroller.lyzr.com/blog/when-to-automatically-quarantine-an-ai-agent.md)
- [What AI agent controls do banks need now that model risk guidance excludes agentic AI?](https://opencontroller.lyzr.com/blog/ai-agent-controls-for-banks-model-risk.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
