Definition
CAISI is the US government's main technical body (housed at NIST) for testing frontier and open-weight AI models for safety and cyber-offensive risk, and for shaping federal AI evaluation standards. It publishes formal capability assessments of specific named models, including foreign-developed ones.
Why it matters
Leadership teams tracking US AI policy should watch CAISI closely — its assessments and leadership stability (it recently lost its director) directly shape which models get flagged as risky and how the US approaches AI export and procurement rules.