NIST opens a four-stage AI evaluation draft

NIST is taking comments through October 6 on a draft four-stage TEVV-Athlon framework for testing, evaluating, verifying and validating AI systems, including agents. The NIST draft and comment notice says the method builds organization-specific assessments from Events and Tools that produce data for measurement Blocks; the 60-day window opened August 7. Operators should treat it as a prompt to write workload-specific acceptance tests before procurement, not as a certification stamp; the archive’s agent-evaluation governance analysis shows why a generic score does not establish deployment safety.