Jailbreak Attack Success on AdvBench GPT-4O-MINI
55.6ASR@1ATOXIA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ATOXIATarget LLM=GPT-4O-MINI, Setting=Black-box2025.10 | 55.6 | 88.3 | |
| ADVPROMPTERTarget LLM=GPT-4O-MINI, Setting=Black-box2025.10 | 35.1 | 79.2 | |
| RED-BANDIT (ϵ-GREEDY)Target LLM=GPT-4O-MINI, Setting=Black-box, Policy=ϵ-greedy2025.10 | 31.2 | 94.2 | |
| RED-BANDIT (UCB)Target LLM=GPT-4O-MINI, Setting=Black-box, Policy=UCB2025.10 | 24 | 92.2 |