Generative AI can help businesses create content, analyze information, write code, and automate routine tasks. But it also introduces risks that traditional software controls may not fully address. An AI assistant might invent legal citations, expose confidential information, or generate misleading content that employees mistake for fact.

 

The NIST AI RMF Generative AI Profile helps identify and manage these risks. Published as NIST AI 600-1 in July 2024, it builds on the NIST AI Risk Management Framework (AI RMF) and applies its approach to generative AI.

 

This guide explains what the profile covers, how its risk categories and suggested actions work, and how organizations can use it in their AI governance programs.

 

 

NIST AI RMF Generative AI Profile at a glance

 

Aspect	Details
Official name	Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile
Document no	NIST AI 600-1
Published	July 26, 2024
Framework it extends	NIST AI RMF 1.0
Risk categories	12
Suggested actions	More than 200
Intended use	Voluntary guidance for managing generative AI risks

 

What is the NIST AI RMF Generative AI Profile?

 

The profile is a companion resource to the NIST AI RMF 1.0. It helps organizations identify and manage risks that are unique to, or made more significant by, generative AI: systems that produce text, images, audio, video, and other content.

 

The NIST AI RMF 1.0, published in January 2023, organizes AI risk management around four functions: Govern, Map, Measure, and Manage. The Generative AI Profile doesn’t create a separate set of functions. It uses the AI RMF’s existing functions, categories, and subcategories to organize generative AI-specific guidance, such as testing for fabricated outputs, evaluating content provenance methods, and reviewing third-party AI components.

 

The profile addresses these concerns through two main components:

 

• Risk categories: Twelve categories describe generative AI risks, from confabulation and data privacy to environmental impacts and third-party component integration.

 

• Suggested actions: More than 200 actions help organizations apply the AI RMF to those risks. They cover activities such as establishing governance practices, testing systems, evaluating content provenance, and managing incidents.

 

The profile does not prescribe one process for every organization. It provides guidance that organizations can adapt to their use cases, risk priorities, resources, and applicable requirements.

 

For example, a business using an AI assistant to draft internal documents faces different risks than a company integrating generative AI into a customer-facing application. Both can use the profile, but their assessments and safeguards should reflect their respective contexts.

 

How the four AI RMF functions apply to generative AI

 

The four functions belong to the AI RMF, not to the profile. The profile connects them to the particular risks and practices of generative AI. Here is what it adds to each.

 

1. Govern

 

Generative AI governance goes beyond approving tools. Organizations need policies for approved tools and use cases, records of AI systems and their third-party dependencies, and defined procedures for reporting, investigating, and disclosing incidents. 

 

For example, a company giving employees an AI assistant with access to internal documents should decide who approves the integration, who sets access permissions, and who investigates an incident involving confidential data.

 

2. Map

 

The same model carries different risks depending on how it’s used. A brainstorming chatbot is not the same as an assistant that retrieves confidential records. Mapping means documenting the system’s purpose and limitations, identifying affected stakeholders and foreseeable misuse, and considering whether AI-generated content needs provenance information.

 

3. Measure

 

Generative AI behavior varies with prompts, context, and configuration, so a successful demo doesn’t prove a system is safe. The profile emphasizes pre-deployment testing and AI red-teaming, where testers deliberately probe for harmful outputs, security weaknesses, and bypassable safeguards. Findings should inform deployment decisions.

 

4. Manage

 

AI risk management doesn’t end at launch. Models, data, integrations, and attack methods change. The profile stresses incident response and disclosure, post-deployment monitoring, and reassessment after significant changes, with documented lessons feeding back into future testing.

 

 

What are the 12 risk categories in the NIST AI RMF Generative AI Profile?

 

NIST AI 600-1 identifies 12 risks that are unique to or exacerbated by generative AI. The categories cover technical failures, misuse, human-AI interactions, third-party dependencies, and wider societal impacts.

 

These categories can overlap. A single incident might involve data privacy, information security, and value chain risks at the same time. Organizations should use the categories to structure their assessments, not assume every incident fits neatly into one category.

 

What are the four cross-cutting themes in the profile?

 

The Generative AI Public Working Group focused on four primary considerations that inform the profile’s suggested actions. These themes are particularly useful for understanding what the profile adds to the broader AI RMF.

 

Governance

 

AI governance establishes how an organization oversees generative AI, assigns responsibility, sets policies, and makes risk-related decisions.

 

The profile includes actions related to organizational accountability, risk management practices, supplier relationships, and procedures for testing and disclosing incidents.

 

For example, a business can establish an approval process for new AI tools, identify who owns each system, and define when security, privacy, or legal teams must review a proposed use case.

 

Content provenance

 

Content provenance concerns the origin and history of digital content and the ability to assess its authenticity.

 

Generative AI makes it easier to create synthetic content that resembles human-produced material. Provenance techniques can help record where content came from, how it changed, and whether relevant authenticity information is available.

 

Depending on the use case, organizations can consider metadata, digital watermarking, cryptographic methods, content labeling, and tools that analyze content authenticity. These techniques have limitations, and no single method establishes the authenticity of every item of content.

 

For example, a business publishing AI-generated media could record its origin and modifications and clearly disclose its use of generative AI where appropriate.

 

Pre-deployment testing

 

Pre-deployment testing evaluates a generative AI system before releasing it for its intended use.

 

The profile emphasizes testing and evaluation, including red-teaming, to identify weaknesses and assess whether safeguards work. Testing should reflect the system’s intended use, relevant stakeholders, and potential consequences of failure.

 

For example, an organization deploying an AI assistant for customer support could test how it handles inaccurate information, sensitive inputs, inappropriate requests, and attempts to bypass safeguards.

 

The findings should inform deployment decisions and any additional work needed before release.

 

Incident disclosure

 

Incident disclosure addresses how organizations identify, document, communicate, and learn from generative AI incidents.

 

Organizations need clear processes for reporting problems, determining their impact, notifying relevant stakeholders where appropriate, and incorporating lessons into risk management.

 

For example, if an AI-enabled service exposes confidential information, the organization should investigate the incident, follow its response procedures, assess disclosure obligations, and determine what changes are needed to prevent recurrence.

 

How are the profile’s 200+ suggested actions organized?

 

The profile includes more than 200 suggested actions to help organizations address the risks it identifies. These actions are organized around relevant subcategories of the broader AI RMF, making it easier to connect generative AI-specific practices with existing risk management activities.

 

Each action has an identifier, a description, and one or more associated generative AI risks. The risk tags help organizations see which concerns an action is intended to address.

 

For example, an action might recommend evaluating the effectiveness of content provenance methods. Another might focus on red-teaming a system to assess its security and resilience. Other actions address supplier assessments, documentation, monitoring, or incident response.

 

This structure provides three practical benefits:

 

• Traceability: Organizations can connect a risk assessment to relevant guidance and document why they selected particular actions.

 

• Prioritization: Teams can focus on actions that address the risks and impacts relevant to their AI systems.

 

• Integration: Organizations can incorporate selected actions into existing governance, security, privacy, and enterprise risk management processes.

 

The actions are not a mandatory checklist, and NIST does not expect every organization to implement all of them in a fixed order. Their relevance depends on the system, its lifecycle stage, the organization’s resources, and the risks involved.

 

How CyberArrow can support AI risk governance

 

Applying AI risk management guidance across an organization requires clear ownership, consistent documentation, and visibility into risk treatment and control effectiveness.

 

CyberArrow helps organizations centralize GRC activities, including risk assessments, control tracking, evidence collection, and audit workflows. These capabilities can support the governance processes used to operationalize AI risk management guidance.

 

With CyberArrow, organizations can:

 

• Document AI-related risks and assign responsibility for managing them.
• Track mitigation plans and monitor remediation activities.
• Maintain evidence of controls and risk management decisions.
• Connect AI risk oversight with broader risk, compliance, and audit processes.

 

If you’re looking to strengthen your AI governance processes, schedule a free demo with CyberArrow to explore how its GRC capabilities can support your organization’s risk management needs.

 

 

FAQs