CAISI Cyber Safeguards - Best Public Jailbreak
Safety · 2026-07-17
The jailbreak condition of CAISI Cyber Safeguards reports fully compliant responses after selecting the most effective of 30 public jailbreaks for each model. Lower scores indicate greater robustness to public jailbreak prompting.
Top models (lower is better)
| Model | Score |
|---|---|
| Opus 4.8 | 7.0 |
| GLM-5.2 | 8.0 |
| Kimi K2.6 | 9.0 |
| GPT-5.5 | 12.0 |
| DeepSeek-V4-Pro | 100.0 |