Jailbreak Defense on Adversarial Jailbreak Attacks (4 Specific Vectors)
100Safety Score (Cipher)Claude Opus 4.5
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Claude Opus 4.5Defense Pipeline=Graph-augmented pedagogical pipeline2026.04 | 100 | 94.37 | 91.55 | 94.37 | |
| Gemini 2.5 Flash-LiteDefense Pipeline=Graph-augmented pedagogical pipeline2026.04 | 94.37 | 89.44 | 92.25 | 91.55 | |
| Gemini 2.5 FlashDefense Pipeline=Graph-augmented pedagogical pipeline2026.04 | 93.66 | 92.96 | 92.96 | 38.03 | |
| GPT-5Defense Pipeline=Graph-augmented pedagogical pipeline2026.04 | 93.66 | 91.55 | 90.85 | 92.25 | |
| Gemini 2.5 ProDefense Pipeline=Graph-augmented pedagogical pipeline2026.04 | 92.96 | 92.96 | 95.07 | 89.44 |