Model Online / On-Site
Duration
Modules 5 modules

Course Objective & Scope

While general AI risk frameworks handle predictive models well, Large Language Models (LLMs), multimodal generators and agents introduce distinct risks such as confabulation (hallucination), copyright infringement, prompt injection and opaque value-chain dependencies. NIST AI 600-1 (the Generative AI Profile) provides over 200 concrete actions specifically designed to govern these risks.

This course unpacks NIST AI 600-1's 12 generative AI risk categories and maps its suggested actions across Govern, Map, Measure and Manage. Participants build a tailored GenAI Risk Control Matrix and TEVV readiness checklist for their organisation's LLM and RAG initiatives.

Who Should Attend?

  • Product and governance leaders overseeing LLM, RAG, copilot and agent initiatives
  • Model Risk Management (MRM), information security and enterprise risk specialists
  • AI CoE teams, enterprise architects and AI evaluation/quality leads
  • Compliance managers extending NIST AI RMF practices to generative AI

Key Takeaways

  • Map NIST AI 600-1's 12 generative AI risk categories to your enterprise LLM and multimodal use cases
  • Select targeted NIST controls for confabulation, data privacy, IP infringement and offensive cyber risks
  • Clarify shared responsibility across foundation model providers, fine-tuners and enterprise deployers
  • Define GenAI-specific TEVV practices including RAG groundedness evals, red teaming and content provenance
  • Produce a mandatory NIST AI 600-1 pre-launch control checklist for internal GenAI projects

Curriculum

Unpacking the 12 Generative AI Risk Categories in NIST AI 600-1
  • Confabulation (hallucination), information integrity, data privacy and intellectual property risks
  • Cybersecurity/prompt attacks, harmful/obscene content, value-chain and component integration risks
  • Human-AI configuration (overreliance/anthropomorphism), compute footprint and CBRN risks
GOVERN and MAP for GenAI: Value Chain, Provenance and Context Limits
  • Allocating accountability across foundation model providers, fine-tuners and enterprise deployers
  • GenAI acceptable use policies, training data provenance checks and copyright safeguards
  • Mapping high-stakes contexts where unverified generative outputs must be restricted
MEASURE for GenAI: Tailored TEVV, Red Teaming and Groundedness
  • Why public LLM leaderboards do not predict enterprise domain performance
  • Measuring confabulation rates, RAG groundedness, stereotype bias and jailbreak resilience
  • Running structured GenAI red-teaming exercises and calibrated human evaluations
MANAGE for GenAI: Guardrails, Content Provenance and Live Controls
  • Grounding via RAG, input/output guardrails and abstention ('I don't know') behaviour
  • Synthetic content provenance (C2PA metadata, watermarking) and user disclosures
  • Handling upstream model drift, prompt regressions and emergent production behaviours
Workshop: Building a NIST AI 600-1 Control Matrix for an Enterprise LLM Use Case
  • Selecting the highest-priority NIST AI 600-1 risk categories for a real internal GenAI project
  • Translating NIST AI 600-1 action IDs into a mandatory pre-production gate checklist

Other Courses in This Category

Get in touch