List your company

Jailbreak Attack Safety Evaluation on CARES adversarial mental-health targets 2025

84Success Score (SS)AMA
34.0847.046072.96Apr 6, 2026

Evaluation results

MethodLinks
AMATarget Model=Llama-3.1-8B2026.048429
Actor-AttackTarget Model=Psycho-8B2026.048034
AMATarget Model=Psycho-8B2026.047738
AMATarget Model=Qwen-2.5-72B2026.047740
Actor-AttackTarget Model=Llama-3.1-70B2026.047741
AMATarget Model=GPT-3.52026.047632
Actor-AttackTarget Model=GPT-3.52026.047641
Actor-AttackTarget Model=Qwen-2.5-72B2026.047640
CoA (new)Target Model=GPT-3.52026.047544
AMATarget Model=Llama-3.1-70B2026.047544
AMATarget Model=Casper-7B2026.047537
Actor-AttackTarget Model=GPT-4o-mini2026.047538
AMATarget Model=Qwen-2-7B2026.047447
Actor-AttackTarget Model=Casper-7B2026.047442
CoA (new)Target Model=Qwen-2.5-72B2026.047349
Actor-AttackTarget Model=Llama-3.1-8B2026.047345
CoA (new)Target Model=Llama-3.1-8B2026.047256
CoA (new)Target Model=Qwen-2-7B2026.047152
CrescendoTarget Model=Qwen-2-7B2026.047051
CoA (new)Target Model=Llama-3.1-70B2026.046959
Actor-AttackTarget Model=Qwen-2-7B2026.046953
CoA (new)Target Model=Psycho-8B2026.046758
AMATarget Model=GPT-4o-mini2026.046748
CrescendoTarget Model=Qwen-2.5-72B2026.046755
CoA (new)Target Model=Casper-7B2026.046653
CrescendoTarget Model=Llama-3.1-8B2026.046669
CrescendoTarget Model=GPT-3.52026.046658
CrescendoTarget Model=GPT-4o-mini2026.046552
CrescendoTarget Model=Casper-7B2026.046263
CrescendoTarget Model=Psycho-8B2026.046262
CrescendoTarget Model=Llama-3.1-70B2026.046164
CoA (new)Target Model=GPT-4o-mini2026.046059
PCSATarget Model=Qwen-2-7B2026.045967
PCSATarget Model=Llama-3.1-70B2026.045377
PCSATarget Model=GPT-4o-mini2026.045374
PCSATarget Model=Qwen-2.5-72B2026.045177
PCSATarget Model=Psycho-8B2026.044686
PCSATarget Model=Llama-3.1-8B2026.044384
PCSATarget Model=Casper-7B2026.044084
PCSATarget Model=GPT-3.52026.043688