Evidence note

NIST AI 600-1: Generative Artificial Intelligence Profile

Government Guidance2024National Institute of Standards and Technology

What does it cover?

A Generative AI profile extending the NIST AI Risk Management Framework with risks and actions particularly relevant to generative AI systems.

Key points

01The profile addresses risks including confabulation, data privacy, information integrity, information security and component integration.
02It recommends evaluating system performance under conditions similar to deployment.
03It recommends reassessing risk when systems are adapted to materially new domains or contexts.
04It emphasises ongoing testing, monitoring, security evaluation and incident management.

Why it matters. Lornets interpretation.

Generic model benchmarks are insufficient evidence for a production AI application. Assurance needs to consider the actual deployment context, integrations, monitoring and change over time.

This is the Lornets reading of the source, not a finding of the source itself.

What it does not establish

01The profile does not certify individual applications.
02It is risk-management guidance rather than a guarantee of safe system behaviour.
03Applicable depth should remain proportionate to context.