Assess AI risk before release

8 steps · about 3 h 55 min · Intermediate

  1. Frame AI risk as an ongoing practice (guide, NIST, 25 min)

    The framework connects governance, context, measurement and management across the life cycle.

  2. Read the AI RMF core (reference, NIST AI Resource Center, 35 min)

    The core gives the four functions and outcomes that will structure the assessment.

  3. Turn outcomes into practical actions (guide, NIST AI Resource Center, 35 min)

    The playbook offers candidate actions without pretending one checklist fits every system.

  4. Compare candidate risk actions (guide, NIST, 30 min)

    The playbook expands the core with practical options that can be selected for this context.

  5. Identify unresolved measurement gaps (reference, NIST, 25 min)

    The roadmap highlights areas where evidence and methods are still developing.

  6. Add generative-AI-specific risks when relevant (guide, NIST, 30 min)

    Generative systems introduce misuse, confabulation and content risks that need explicit treatment.

  7. Define the trustworthiness characteristics to measure (reference, NIST AI Resource Center, 25 min)

    Valid, safe, transparent and fair are separate characteristics that need concrete evidence.

  8. Document intended use and evaluation (guide, Google, 30 min)

    A model card makes limitations, metrics and decisions inspectable after release.