NIST published a first public draft of a framework for testing and evaluating AI systems, numbered AI 200-2 and named TEVV-Athlon. It sets out a four-stage method organisations can use to build assessments tailored to their own goals, covering conventional machine learning models, language models and agent-based systems. NIST is taking written comments until 6 October 2026, and says all submissions may be released under freedom of information rules.
What changed
NIST's risk management framework called for testing and evaluation of AI systems without setting out a structured method for building one.
What it unlocks
Following a published step-by-step method to design a customised assessment of an AI system, and commenting on it before it is finalised.
- comment period closes 6 October 2026
- 60-day comment period
- four-stage assessment method
What you need to act on it
- comments emailed to NIST by 6 October 2026 with "NIST AI 200-2" in the subject line
Sources