The AI RMF 1.0 (January 2023) is the umbrella framework, four functions (Govern, Map, Measure, Manage) that apply to any AI system, classical ML included. The Generative AI profile (NIST AI 600-1, July 2024) is a companion document that takes those four functions and adds the risks specific to generative systems: confabulation, data leakage from training data, prompt injection, dangerous-content generation, value-chain risks, and so on. An LLM product touches both. You inherit the RMF structure, then you actually apply the GenAI profile because that is where the LLM-specific risks and controls are listed. Citing only the umbrella RMF in a customer questionnaire is a tell that the team has not done the LLM-specific work.
NIST published two documents teams confuse with each other.
The **AI Risk Management Framework 1.0** came out in January 2023. It is the umbrella framework: four functions called Govern, Map, Measure, and Manage, applicable to any AI system. It is technology-agnostic, the same structure covers a classical fraud-detection model, a recommendation system, and a large language model.
The **Generative AI profile** (NIST AI 600-1) came out in July 2024. It is a companion document. It does not replace the RMF. Instead, it takes the four functions and overlays the risks specific to generative systems: confabulation (hallucination), training-data leakage, prompt injection, dangerous-content generation, value-chain risks (training data licensing, foundation-model dependencies), CBRN risks at the high end.
An LLM product is subject to both. You inherit the RMF's structure, the four functions, the Govern/Map/Measure/Manage cycle, and then you actually apply the GenAI profile because that is where the LLM-specific controls live.
The tell that someone has not done the work: they cite the umbrella RMF and never mention the GenAI profile. That means they have the framework structure but have not engaged with the LLM-specific risks.
**Two documents, both apply, in this order.**
The **AI RMF 1.0** is the foundational document. Four functions: Govern (policies, roles, accountability), Map (context, stakeholders, intended uses), Measure (testing, monitoring, metrics), Manage (treatment, monitoring, response). It is intentionally technology-agnostic, a single page might describe controls that apply equally to a fraud-detection model and a chatbot.
The **Generative AI profile (NIST AI 600-1)** is the LLM-specific overlay. It enumerates the risks particular to generative systems, confabulation, data leakage, environmental impact, dangerous content, value-chain dependencies, intellectual property, prompt injection, information integrity, and maps each to specific controls under the four RMF functions. It does not replace the RMF; it tells you what concretely to do within the RMF's structure when the system happens to be generative.
An LLM product needs both. The RMF gives you the governance scaffold; the GenAI profile gives you the LLM-specific risk register and control library. A vendor questionnaire that names only the RMF without referencing the GenAI profile signals that the team has the structure but has not done the LLM-specific control mapping, and that is the substantive work the buyer cares about.
The interview signal is knowing both documents exist, knowing the RMF is the umbrella and the GenAI profile is the LLM specifics, and recognising that 'we use the NIST framework' is a half-answer without the profile name attached.
Quick reference
TL;DR
AI RMF 1.0 is the umbrella four-function framework for any AI system; the GenAI profile (NIST AI 600-1) adds the LLM-specific risks like injection, confabulation, and data leakage on top.
Imagine a general building code that covers every type of structure, houses, schools, hospitals. Then imagine a supplementary code that only applies to hospitals, listing risks unique to hospitals like infection control and oxygen handling. A hospital follows both. The AI RMF is the general code; the Generative AI profile is the hospital-specific supplement. If someone tells you their hospital follows only the general code, you know they have not done the hospital-specific safety work.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example.
Your product is a non-generative classical ML system (fraud detection, image classification) The GenAI profile's specific risks (confabulation, prompt injection) do not apply; the RMF alone is the appropriate reference
You operate exclusively under a non-NIST regime (EU AI Act with no US procurement exposure, sectoral regimes like FDA SaMD) NIST is not a binding standard; the time spent mapping controls might be better invested in the actually-binding framework
Pre-launch internal prototype with no external users or customer commitments Formal control mapping is overhead the prototype phase does not benefit from; start when the product begins to face external compliance asks
Quick scan · 2–4 min
Everything important, quickly.
6 min: name both documents, explain the umbrella + specific layering, walk the GenAI risk categories, describe how to use both together, and contextualise against ISO 42001 and the EU AI Act.
Side-by-side
Aspect
AI RMF 1.0
GenAI Profile (NIST AI 600-1)
Published
January 2023
July 2024
Scope
Any AI system, including classical ML
Generative AI specifically
Structure
Four functions: Govern, Map, Measure, Manage
Risk categories + control mapping into the four functions
Names risks like prompt injection
No
Yes
Applies to an LLM product
Yes, as the umbrella structure
Yes, as the LLM-specific control layer
Real-world examples
Real products, models, and research that use this idea.
1Anthropic's Responsible Disclosure framework references the GenAI profile risks (confabulation, dangerous content, CBRN uplift) explicitly in their public safety documentation.
2OpenAI's enterprise customer trust portal references AI RMF 1.0 alignment and breaks out GenAI-profile-specific controls in the appendix.
3Microsoft's Responsible AI Standard v3 maps directly to the four RMF functions and references the GenAI profile risk categories.
4AWS Bedrock's compliance documentation (2026) ships an RMF-aligned controls matrix and a separate GenAI-profile-aligned controls matrix for customers.
Interviewer follow-ups
What an interviewer would ask next. Try answering before peeking at the approach.
QHow does the GenAI profile relate to ISO 42001 and the EU AI Act?
Sign in or go Pro to see the approach for this follow-up.
QWhat changes when NIST updates the GenAI profile?
Sign in or go Pro to see the approach for this follow-up.
Don't say this
Red flags and common mistakes that signal junior thinking. Click to expand.
Most common mistake
Citing only the AI RMF 1.0 in a compliance answer for an LLM product. The GenAI profile is where the LLM-specific risks and controls actually live.