Zhipu AI GLM-4.6: LLM Security & Jailbreak Resistance
As of 2026-07-04, Zhipu AI GLM-4.6 has an attack success rate (ASR) of 39.0% on Guardion's LLM Vulnerability benchmark — ranking #53 of 93 models. It is weakly resistant — adversarial prompts succeed often without added guardrails.
Attack Success Rate (ASR) is the share of adversarial prompts that elicit harmful output across zero-shot, TAP, and Crescendo attacks (HarmBench framework). Lower is safer. "Est." models are calibrated from public safety evaluations pending a Guardion benchmark run.
How GLM-4.6 compares
Frequently asked questions
How secure is Zhipu AI GLM-4.6 against jailbreaks?
Zhipu AI GLM-4.6 is weakly resistant — adversarial prompts succeed often without added guardrails. On Guardion's LLM Vulnerability benchmark it records a 39.0% attack success rate (61% robustness), ranking #53 of 93 models (estimated from public safety evaluations).
What is GLM-4.6's attack success rate (ASR)?
GLM-4.6's overall ASR is 39.0% — the share of adversarial prompts that elicit harmful output across zero-shot, TAP, and Crescendo attacks. Lower is safer.
Is GLM-4.6 more secure than GLM-4.5?
GLM-4.6 is more robust: 39.0% ASR vs GLM-4.5's 41.0%. See the full head-to-head comparison for the per-attack breakdown.
How can I make GLM-4.6 safer to deploy?
Even robust models are bypassed under sustained attack. An inline runtime guardrail that classifies prompts and responses — like Guardion's prompt-defense models — blocks jailbreaks and prompt injection before they reach GLM-4.6.
Harden GLM-4.6 in production
Guardion's runtime guardrails block jailbreaks and prompt injection before they reach any model — with sub-130ms policy decisions.
Request a demo