Jailbreak Attack Success on AdvBench GPT-3.5-TURBO
61.4ASR@1ATOXIA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ATOXIATarget LLM=GPT-3.5-TURBO, Setting=Black-box2025.10 | 61.4 | 92.7 | |
| ADVPROMPTERTarget LLM=GPT-3.5-TURBO, Setting=Black-box2025.10 | 43.5 | 86 | |
| RED-BANDIT (ϵ-GREEDY)Target LLM=GPT-3.5-TURBO, Setting=Black-box, Policy=ϵ-greedy2025.10 | 35.6 | 97.1 | |
| RED-BANDIT (UCB)Target LLM=GPT-3.5-TURBO, Setting=Black-box, Policy=UCB2025.10 | 30.8 | 98.1 |