Jailbreak Attack on AdvBench GPT-4o
6ASRGCG (transfer)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GCG (transfer)Attack setting=Transferred2025.12 | 6 | — | — | |
| DoublespeakAttack setting=Optimization-free, Zero-shot=true2025.12 | 31 | — | — | |
| Adaptive (transfer)Attack setting=Transferred2025.12 | 42 | — | — | |
| PAIRAttack setting=Optimization-based2025.12 | 62 | — | — | |
| TAPAttack setting=Optimization-based2025.12 | 88 | — | — | |
| Adv. ReasoningAttack setting=Optimization-based2025.12 | 94 | — | — | |
| FlipAttackAttack setting=Optimization-free2025.12 | 98 | — | — | |
| ADVPROMPTERTarget LLM=GPT-4O, Setting=Black-box2025.10 | — | 27.8 | 68.9 | |
| ATOXIATarget LLM=GPT-4O, Setting=Black-box2025.10 | — | 48.2 | 82.4 | |
| RED-BANDIT (UCB)Target LLM=GPT-4O, Setting=Black-box, Policy=UCB2025.10 | — | 29.8 | 93.3 | |
| RED-BANDIT (ϵ-GREEDY)Target LLM=GPT-4O, Setting=Black-box, Policy=ϵ-greedy2025.10 | — | 36.6 | 85.6 |