Jailbreak Attack Safety Evaluation on CARES adversarial mental-health targets 2025
84Success Score (SS)AMA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AMATarget Model=Llama-3.1-8B2026.04 | 84 | 29 | |
| Actor-AttackTarget Model=Psycho-8B2026.04 | 80 | 34 | |
| AMATarget Model=Psycho-8B2026.04 | 77 | 38 | |
| AMATarget Model=Qwen-2.5-72B2026.04 | 77 | 40 | |
| Actor-AttackTarget Model=Llama-3.1-70B2026.04 | 77 | 41 | |
| AMATarget Model=GPT-3.52026.04 | 76 | 32 | |
| Actor-AttackTarget Model=GPT-3.52026.04 | 76 | 41 | |
| Actor-AttackTarget Model=Qwen-2.5-72B2026.04 | 76 | 40 | |
| CoA (new)Target Model=GPT-3.52026.04 | 75 | 44 | |
| AMATarget Model=Llama-3.1-70B2026.04 | 75 | 44 | |
| AMATarget Model=Casper-7B2026.04 | 75 | 37 | |
| Actor-AttackTarget Model=GPT-4o-mini2026.04 | 75 | 38 | |
| AMATarget Model=Qwen-2-7B2026.04 | 74 | 47 | |
| Actor-AttackTarget Model=Casper-7B2026.04 | 74 | 42 | |
| CoA (new)Target Model=Qwen-2.5-72B2026.04 | 73 | 49 | |
| Actor-AttackTarget Model=Llama-3.1-8B2026.04 | 73 | 45 | |
| CoA (new)Target Model=Llama-3.1-8B2026.04 | 72 | 56 | |
| CoA (new)Target Model=Qwen-2-7B2026.04 | 71 | 52 | |
| CrescendoTarget Model=Qwen-2-7B2026.04 | 70 | 51 | |
| CoA (new)Target Model=Llama-3.1-70B2026.04 | 69 | 59 | |
| Actor-AttackTarget Model=Qwen-2-7B2026.04 | 69 | 53 | |
| CoA (new)Target Model=Psycho-8B2026.04 | 67 | 58 | |
| AMATarget Model=GPT-4o-mini2026.04 | 67 | 48 | |
| CrescendoTarget Model=Qwen-2.5-72B2026.04 | 67 | 55 | |
| CoA (new)Target Model=Casper-7B2026.04 | 66 | 53 | |
| CrescendoTarget Model=Llama-3.1-8B2026.04 | 66 | 69 | |
| CrescendoTarget Model=GPT-3.52026.04 | 66 | 58 | |
| CrescendoTarget Model=GPT-4o-mini2026.04 | 65 | 52 | |
| CrescendoTarget Model=Casper-7B2026.04 | 62 | 63 | |
| CrescendoTarget Model=Psycho-8B2026.04 | 62 | 62 | |
| CrescendoTarget Model=Llama-3.1-70B2026.04 | 61 | 64 | |
| CoA (new)Target Model=GPT-4o-mini2026.04 | 60 | 59 | |
| PCSATarget Model=Qwen-2-7B2026.04 | 59 | 67 | |
| PCSATarget Model=Llama-3.1-70B2026.04 | 53 | 77 | |
| PCSATarget Model=GPT-4o-mini2026.04 | 53 | 74 | |
| PCSATarget Model=Qwen-2.5-72B2026.04 | 51 | 77 | |
| PCSATarget Model=Psycho-8B2026.04 | 46 | 86 | |
| PCSATarget Model=Llama-3.1-8B2026.04 | 43 | 84 | |
| PCSATarget Model=Casper-7B2026.04 | 40 | 84 | |
| PCSATarget Model=GPT-3.52026.04 | 36 | 88 |