NIST Generative AI Profile

What is the NIST Generative AI Profile?

The NIST Generative AI Profile, formally NIST AI 600-1, was published in July 2024 as a companion to AI RMF 1.0. It is one of an expanding family of “Profiles” within the AI RMF ecosystem that overlay the general framework with use-case-specific or risk-domain-specific guidance.

The GenAI Profile catalogs twelve risks that NIST identified as either unique to generative AI or significantly amplified by generative AI compared with classical machine learning. Among the twelve: confabulation (hallucination), dangerous or violent content recommendations, data privacy leakage, environmental cost, harmful bias and homogenization, human-AI configuration, information integrity, information security, intellectual property, obscene or indecent content, value chain and component integration risk, and chemical, biological, radiological, or nuclear (CBRN) information.

For each risk, the Profile maps actions across the four AI RMF functions (Govern, Map, Measure, Manage) and notes whether the action falls on the developer, the deployer, or both. The Profile is not prescriptive about which actions an organization must take; it is a structured menu.

Why does this matter for law firms?

Of the twelve risks, four are directly relevant to attorney AI use and are the substantive content of most state bar AI opinions and ABA Formal Opinion 512:

Confabulation (hallucination). The fabrication of plausible-sounding but factually wrong output. The hallucination risk is the throughline of every published US sanctions case against an attorney for AI use, beginning with Mata v. Avianca (S.D.N.Y. 2023). The GenAI Profile’s measurement actions for confabulation map onto the verification requirements in attorney candor duties (Model Rule 3.3).

Data privacy leakage. Inputs to generative AI systems may surface in subsequent outputs. ABA Op 512 and state bar opinions on confidentiality (Model Rule 1.6) treat this as the controlling concern. The Profile’s Govern actions for data minimization and Manage actions for prompt-handling map to firm-level confidentiality policy.

Intellectual property. The Profile addresses both training-data IP and output IP. For attorneys, the issue is bidirectional: client work-product confidentiality on the input side, and ownership and infringement risk on the output side.

Information integrity / information security. Adjacent to confidentiality but framed around system integrity rather than data leakage. Relevant to firm vendor due diligence under Model Rule 1.6 and to incident response.

How does this relate to AI RMF 1.0?

The GenAI Profile sits inside AI RMF 1.0; it does not replace it. An organization that has adopted AI RMF and is then deploying generative AI applies the GenAI Profile as the overlay specifying which Map, Measure, and Manage actions are most material for the GenAI risk catalog.

The relationship is orthogonal to the framework’s voluntary status. NIST does not certify either AI RMF or the GenAI Profile. Adoption is signaled through documentation: a firm AI policy that names AI RMF and the GenAI Profile as the structuring frameworks, then traces specific firm controls (verification protocols, vendor diligence, incident response) to the corresponding Profile actions, is doing the demonstrable work that bar counsel, regulators, and carriers are looking for.

Does this survive EO 14110’s revocation?

The GenAI Profile was developed pursuant to direction in Executive Order 14110 (October 2023, signed by President Biden). EO 14110 was revoked by Executive Order 14179 (January 2025, signed by President Trump). NIST publications stand on their own as agency publications and do not require an active Executive Order to remain in force. The GenAI Profile remains available, citable, and operationally relevant. NIST has not announced rescission or replacement.

Primary sources

• NIST, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1 (July 2024): PDF on nvlpubs.nist.gov.