Assess AI risk before release
8 steps · about 3 h 55 min · Intermediate
- Frame AI risk as an ongoing practice (guide, NIST, 25 min)
The framework connects governance, context, measurement and management across the life cycle.
- 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.
- Turn outcomes into practical actions (guide, NIST AI Resource Center, 35 min)
The playbook offers candidate actions without pretending one checklist fits every system.
- Compare candidate risk actions (guide, NIST, 30 min)
The playbook expands the core with practical options that can be selected for this context.
- Identify unresolved measurement gaps (reference, NIST, 25 min)
The roadmap highlights areas where evidence and methods are still developing.
- Add generative-AI-specific risks when relevant (guide, NIST, 30 min)
Generative systems introduce misuse, confabulation and content risks that need explicit treatment.
- 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.
- Document intended use and evaluation (guide, Google, 30 min)
A model card makes limitations, metrics and decisions inspectable after release.