US CAISI publishes assessment of Z.ai's GLM-5.2

The US Center for AI Standards and Innovation (CAISI), part of NIST, published an assessment of GLM-5.2, the open-weight model released by the Chinese company Z.ai (formerly Zhipu) in June 2026. CAISI judged that GLM-5.2 “was probably the most capable open-weight AI model when it was released,” finding its overall capabilities broadly comparable to OpenAI’s GPT-5.2 from December 2025, and its cyber capabilities similar to Anthropic’s Opus 4.6 from February 2026.

CAISI’s evaluation of the model’s safeguards was mixed. It found GLM-5.2’s protections would assist with developing agentic cyber exploits, and that it blocked fewer sensitive biology-related questions than reference US models. At the same time, CAISI found it more resistant to agent-hijacking and jailbreaking attacks than other Chinese open-weight models it had evaluated. The assessment noted the standard caveat for any open-weight release: safeguards built into a model can be removed or bypassed once it is self-hosted, so guardrails present in an official deployment do not necessarily persist downstream.

The report was one of a series of US and UK government technical assessments of frontier Chinese open-weight releases published through mid-2026, treating capability parity with closed Western models as now largely established and shifting the focus of government scrutiny toward safeguard robustness.