Skip to content

Core, Profiles and Playbook

The NIST AI RMF is designed for flexible, risk-based use. Its parts serve different purposes and should not be collapsed into a single checklist or maturity score.

NIST AI 100-1 provides:

  • A framing of AI risk and trustworthiness.
  • Intended audiences and AI actors.
  • The four-function AI RMF Core.
  • The concept of Profiles for tailoring the framework.

The framework is voluntary and outcome-oriented. Organisations apply it according to their role, resources, risk tolerance, use case and operating context.

The Core contains:

  • 4 functions
  • 19 categories
  • 72 subcategories

The subcategories express risk-management outcomes. They do not constitute an ordered checklist, and NIST does not assign a universal implementation score to them.

FunctionRelationship
GOVERNCross-cutting conditions, accountability and culture that inform the other functions
MAPContext and risk identification
MEASUREAnalysis, evaluation, testing and monitoring
MANAGEPrioritisation, treatment, response, recovery and communication

A Profile tailors the Core to a particular application, sector, technology, organisation or system. Profiles help describe what is relevant and how outcomes should be prioritised.

The Current Profile describes what is presently achieved for the selected system and context.

It should answer:

  • What practice exists now?
  • Is it specific to this system?
  • Who owns it?
  • Is it documented?
  • Is it operating?
  • Has effectiveness been tested?
  • What evidence or adverse finding affects the conclusion?

The Target Profile describes the desired risk-management position.

It should reflect:

  • Intended purpose and deployment context.
  • Organisational risk tolerance.
  • Legal, contractual and policy requirements.
  • Potential impacts on people, organisations, society and the environment.
  • Technical and operational dependencies.
  • Lifecycle stage.
  • Resource and implementation constraints.
  • Relevant stakeholder expectations.

The difference between Current and Target Profiles becomes a prioritised improvement plan. A gap may require:

  • A new or strengthened practice.
  • Better ownership or documentation.
  • Additional evidence.
  • A test or evaluation.
  • Supplier action.
  • A deployment restriction.
  • Risk treatment or acceptance.
  • Monitoring and reassessment.

Gamut uses plain assessment language:

Current outcomeMeaning
Not assessedNo defensible determination has been made.
Not achievedThe outcome is absent or materially ineffective for this system.
Partially achievedSome elements operate, but material gaps remain.
AchievedThe outcome operates for this system and is sufficiently supported.

The Target Profile uses the substantive target outcomes: Not achieved, Partially achieved or Achieved.

These are Gamut assessment labels, not a NIST maturity scale.

The NIST AI RMF Playbook provides suggested actions aligned to Core subcategories. NIST describes it as a voluntary companion resource, not a checklist or mandatory sequence.

Use the Playbook to:

  • Explore possible implementation actions.
  • Identify documentation and evidence ideas.
  • Tailor practices to role, sector and use case.
  • Compare alternative ways to achieve an outcome.
  • Support workshops and improvement planning.

Do not:

  • Treat every suggestion as mandatory.
  • Assume one suggestion is sufficient in every context.
  • Copy a suggested action without linking it to the selected system.
  • Represent a Playbook action as evidence that the outcome is achieved.

NIST AI 600-1 is a cross-sectoral companion Profile for generative AI. It identifies generative-AI risk considerations and recommended actions that supplement the AI RMF Core.

When a selected system has generative-AI characteristics, Gamut highlights the Generative AI Profile as an additional lens. The Core remains active; the companion Profile does not replace it.

See Generative AI Profile.

Tailoring should be explicit. Record:

  • The system and context.
  • The outcome being considered.
  • The reason for its priority.
  • The intended target.
  • Any decision to defer work.
  • The accountable decision-maker.
  • The trigger for revisiting the decision.

Tailoring does not mean:

  • Quietly removing inconvenient outcomes.
  • Treating low priority as not applicable.
  • Reducing evidence because a system is familiar.
  • Using another system’s assessment.
  • Ignoring an outcome because a mapped framework appears stronger.
ConceptIt isIt is not
Core outcomeA NIST risk-management outcomeA prescriptive technical control
Current ProfileAssessed present positionA general organisational aspiration
Target ProfileIntended future positionProof of implementation
Playbook actionA voluntary implementation suggestionA mandatory checklist item
Gamut priorityAssessment-planning emphasisA NIST severity rating
Assurance depthStrength of support for a claimThe Current outcome itself
CrosswalkA traceability aidAutomatic equivalence