NIST AI RMF Generative AI Profile (NIST AI 600-1)

What it is

The Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, published as NIST AI 600-1, is a cross-sectoral profile of the NIST AI RMF. NIST released a public draft in April 2024 and the final profile on 26 July 2024, drawing on a public working group of more than 2,500 members.

A profile applies the AI RMF to a particular technology, sector or use. This one takes the AI RMF’s functions and subcategories and asks what they mean for generative AI, the class of models that produce text, images, audio, video and code.

Who it applies to

The profile is written for any organisation that designs, develops, deploys, operates or uses generative AI. That includes model developers, but most of its readers are organisations building applications on third-party models or rolling out generative AI tools to their staff and customers. It applies by choice, or where a customer, contract or law gives it weight.

What it requires

The profile is guidance, so it describes risks and suggested actions rather than mandatory controls.

The 12 risks

The profile describes 12 risks that are unique to generative AI or made worse by it.

• CBRN information or capabilities, meaning easier access to chemical, biological, radiological or nuclear weapons knowledge.
• Confabulation, the confident production of false or misleading content.
• Dangerous, violent or hateful content.
• Data privacy, including leakage and inference of personal data.
• Environmental impacts from the compute used to train and run models.
• Harmful bias and homogenisation of outputs.
• Human-AI configuration, including over-reliance and automation bias.
• Information integrity, such as content that blurs fact and fiction at scale.
• Information security, including lower barriers to cyber attack and new attack surfaces such as prompt injection.
• Intellectual property, such as reproduction of copyrighted or trademarked material.
• Obscene, degrading and abusive content.
• Value chain and component integration, the risk introduced by third-party models, data and components.

Suggested actions

Just over 200 suggested actions are mapped onto the AI RMF’s Govern, Map, Measure and Manage subcategories. Each action is tagged with the risks it addresses and the AI actors it concerns. The profile also gives particular attention to four considerations that run through the actions, which are governance, content provenance, pre-deployment testing and incident disclosure.

Typical actions include setting acceptable use policies for generative AI, reviewing the terms and documentation of third-party models, red-teaming before deployment, monitoring outputs in production, tracking content provenance and setting up channels to report and respond to incidents.

Using it in practice

Few organisations apply all 200 actions. The usual approach is to start from the 12 risks, decide which are material for each generative AI use, and pick the actions that address them. A customer-facing assistant raises confabulation, information integrity and information security risks, so pre-deployment testing, output monitoring and prompt injection defences come first. An internal coding assistant raises intellectual property and value chain risks, so licence terms, data handling and vendor review matter more. Staff use of public chatbots raises data privacy risk, which acceptable use rules and controls on what data leaves the organisation address.

Because many generative AI tools arrive through existing software and subscriptions rather than formal projects, the first practical step is often finding out which generative AI is already in use before the profile can be applied to it.

Key dates

The initial public draft was released in April 2024 and the final profile on 26 July 2024. It remains the current version. NIST has said the AI RMF itself is being revised as part of the White House AI Action Plan, which may in time affect its profiles.

Texas’s Responsible Artificial Intelligence Governance Act (HB 149), in force from 1 January 2026, provides that a defendant may not be found liable if it discovers a violation through an internal review process while substantially complying with this profile or another nationally or internationally recognised AI risk management framework.

How it relates to other frameworks

The profile sits under the NIST AI RMF and uses its structure, so it fits wherever the AI RMF is already in use, including inside an ISO/IEC 42001 management system. Its themes overlap with the G7 Hiroshima Code of Conduct and the EU General-Purpose AI Code of Practice, which address similar risks from the model developer’s side. For organisations that deploy generative AI, the profile is often the most practical source of actions for acceptable use, third-party review and output monitoring.