NIST CAISI publishes capability/cyber-risk assessment of Z.ai's GLM-5.2 open-weight model

NIST's Center for AI Standards and Innovation (CAISI) published an assessment of Z.ai's GLM-5.2 open-weight model on 2026-07-17 (confirmed via json-ld datePublished metadata on nist.gov), evaluating its overall and cyber-offensive capabilities relative to other frontier models (GPT-5.2, Claude Opus 4.6). The assessment applies CAISI's standing evaluation methodology to a newly released open-weight model.

CAISI assessments operationalize the US government's approach to benchmarking frontier/open-weight model risk, including cyber-offensive capability — a reference point AI security teams and policymakers use to gauge which models require heightened deployment safeguards. This is a recurring evaluation activity (not a new framework) but demonstrates active application of the government's AI risk assessment methodology to a specific, newly released model.

Security teams evaluating or deploying GLM-5.2 or comparable open-weight models should review CAISI's cyber-capability findings to calibrate red-teaming and access-control requirements.