NIST AI Risk Management Framework (AI RMF 1.0)

· About NIST AI Risk Management Framework (AI RMF 1.0)

Key Takeaways

  • NIST AI RMF 1.0 is a voluntary U.S. guidance framework, not a binding regulation, and it applies to organizations designing, developing, deploying, or using AI systems.[1][7]
  • It was released on 26 January 2023 as NIST AI 100-1, and NIST states in 2026 that the framework is being revised, with a revised version in progress.[3][6][11]
  • The framework is intended to improve AI trustworthiness by organizing risk management around the functions govern, map, measure, and manage.[1][15]
  • Because the AI RMF is voluntary, it carries no statutory fines or direct enforcement penalties, but NIST profiles and companion resources have become influential baseline controls in procurement, governance, and assurance programs.[1][10][15]
  • NIST’s AI RMF Playbook is a living companion resource that NIST says will be updated after the framework is revised, with updates released about twice per year.[5]
  • A 2024 NIST profile for generative AI exists as a companion resource, showing that the framework is being operationalized through sector-neutral and use-case-specific profiles rather than through mandatory amendments.[10]

What It Is

The NIST AI Risk Management Framework (AI RMF 1.0) is a voluntary, cross-sector guidance document for organizations that design, develop, deploy, or use AI systems to help manage risks to individuals, organizations, and society.[1][7][9] NIST published it as NIST AI 100-1 in January 2023, and the launch page states it was released on 26 January 2023.[3][4][9]

The framework is maintained by the National Institute of Standards and Technology (NIST), a U.S. Department of Commerce agency, and NIST hosts the companion AI Resource Center and AI RMF resources.[1][12][15] NIST’s current 2026 materials state that AI RMF 1.0 is being updated and that a revised version is in progress.[6][11][15]

There are no statutory “in force” or phase-in dates because the AI RMF is not a law or regulation; it is a voluntary framework.[1][7][9] The relevant publication milestone is 26 January 2023, with later companion materials including the July 2024 Generative AI Profile and 2025–2026 resource updates indicating active maintenance rather than legal amendment.[10][11][15]

Who Must Comply

There are no mandatory applicability thresholds in the legal sense because the AI RMF does not compel compliance.[1][7] NIST describes the framework as intended for voluntary use by organizations designing, developing, deploying, or using AI systems.[1][9]

The framework has broad extraterritorial practical reach because any organization outside the United States may adopt it for internal governance, procurement, assurance, or alignment with customer and regulator expectations.[1][10][15] That said, it does not create jurisdictional obligations on non-U.S. entities by itself.[1][7]

There are no formal exemptions because there is no statutory compliance regime to exempt from.[1][7] The main limitation is that the framework is guidance, so organizations may use it selectively, as a baseline, or together with sector rules and contractual obligations.[1][10]

Core Requirements

  1. Establish AI governance. Organizations should define accountability, roles, policies, risk tolerance, and oversight for AI across the lifecycle, because the framework’s first function is govern.[15]
  2. Map AI context and impact. Organizations should identify intended use, stakeholders, harms, benefits, data dependencies, and operational context before deployment, because the framework’s second function is map.[15]
  3. Measure AI risks. Organizations should evaluate performance, bias, robustness, explainability, privacy, and security using appropriate methods and metrics, because the framework’s third function is measure.[1][15]
  4. Manage AI risks continuously. Organizations should prioritize mitigation, monitor residual risks, and maintain incident response and improvement loops, because the fourth function is manage.[1][15]
  5. Use companion profiles and playbooks where relevant. Organizations should adapt the framework through use-case-specific profiles, such as the generative AI profile, rather than treating the core document as a finished control catalog.[10][15]
  6. Treat trustworthiness as lifecycle-wide. NIST positions trustworthiness considerations as relevant to design, development, use, and evaluation, not just pre-release review.[1][9]

Deadlines and Penalties

| milestone | date | what applies | |---|---:|---| | AI RMF 1.0 released | 26 January 2023 | Voluntary framework published by NIST as NIST AI 100-1.[3][4][9] | | Generative AI Profile published | July 2024 | Companion implementation profile for generative AI use cases.[10] | | NIST AI RMF being revised | 2026 | NIST states a revised version is in progress.[6][11][15] |

Because the AI RMF is voluntary, there are no maximum fines, administrative penalties, or criminal sanctions under the framework itself.[1][7][9] Any penalties would arise only if an organization’s separate legal obligations, contracts, procurement terms, or sector-specific laws incorporate AI RMF concepts by reference.

How to Comply

  1. Assign ownership and governance. Create an AI governance committee, define accountable executives, and document decision rights for model approval, release, and retirement.
  2. Inventory AI systems and use cases. Build a register covering model purpose, owner, training data, suppliers, affected populations, and jurisdictions.
  3. Adopt a control baseline. Map governance and risk controls to ISO 27001 for security management, NIST CSF 2.0 for enterprise risk structure, and ISO 42001 for AI management-system discipline where those standards fit the organization’s operating model.
  4. Run pre-deployment risk assessments. Test for bias, robustness, privacy leakage, explainability gaps, prompt-injection exposure, and failure modes before production release.
  5. Document intended use and limitations. Publish internal and external use constraints, human-oversight requirements, escalation triggers, and unacceptable uses.
  6. Implement monitoring and incident response. Track drift, performance degradation, abuse, and harmful outputs; integrate AI events into security and operational incident handling.
  7. Use profiles for higher-risk use cases. Apply the generative AI profile and any sector-specific NIST profiles to tailor controls for sensitive deployments.[10][15]
  8. Review and improve on a fixed cadence. Reassess controls at least annually and after material model, data, vendor, or regulatory changes, consistent with NIST’s living-resource approach.[5][15]

Related Regulations

  • EU AI Act. The AI RMF is a voluntary governance framework, while the EU AI Act is a binding regulation with risk-tiered legal obligations and penalties.
  • NIST Cybersecurity Framework 2.0. The CSF overlaps on governance, risk assessment, and monitoring, but it is aimed at cybersecurity rather than AI-specific trustworthiness.
  • ISO/IEC 42001. ISO 42001 overlaps closely on AI management systems and can operationalize the AI RMF inside an auditable management-system structure.
  • ISO/IEC 27001. ISO 27001 complements the AI RMF on information-security controls, supplier risk, access control, and incident response, especially where AI systems process sensitive data.
  • Sector rules such as HIPAA or financial-services model-risk standards. These can impose binding requirements that the AI RMF helps organize, but they control over the framework if there is a conflict.

FAQ

Does NIST AI RMF 1.0 apply to companies outside the United States?

Yes, in practice it can be used by companies anywhere because it is a voluntary framework for organizations that design, develop, deploy, or use AI systems.[1][7][9] It does not itself impose jurisdictional obligations on non-U.S. firms. International companies often use it as a common governance baseline for procurement, assurance, and customer commitments.[10][15]

Is NIST AI RMF 1.0 legally mandatory?

No. NIST explicitly describes the framework as intended for voluntary use.[1][7][9] Organizations may still be expected to align with it indirectly if a contract, public-sector procurement, or sector program incorporates it.

What changed in 2025–2026?

NIST’s 2026 materials say the AI RMF 1.0 is being updated and that a revised version is in progress.[6][11][15] NIST also maintains the AI RMF Playbook as a living resource and says it will be updated after the framework is revised.[5] No binding amendments or legal delays apply because the framework is not a regulation.

What penalties apply for non-compliance?

There are no penalties built into the framework, because it is voluntary guidance rather than law.[1][7][9] However, failure to follow AI RMF-aligned controls can increase exposure under other laws, contractual warranties, procurement rules, or negligence and consumer-protection theories.

How does the AI RMF relate to generative AI?

NIST published a Generative AI Profile in July 2024 as a companion resource to implement the framework for generative AI systems.[10] That profile does not replace the core framework; it translates the core functions into more specific risk considerations for generative models.

Does the AI RMF replace ISO 27001 or ISO 42001?

No. The AI RMF is a risk-management framework, while ISO 27001 and ISO 42001 are formal management-system standards with auditable requirements. The strongest programs use the AI RMF to shape AI-specific governance and use ISO standards to make the controls operational and certifiable where needed.

Sources

Put it into practice

More compliance guides