Jailbreak Attack on UnsafeDiff
1.7Attack Success Rate (ASR)Ring-A-Bell
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ring-A-BellAttack Protocol=Re-use attack, Safety Filter=OpenAI input/output moderators, Memory Manager=BufferMem2025.04 | 1.7 | 0.216 | 0.535 | — | |
| Ring-A-BellAttack Protocol=One-time attack, Safety Filter=OpenAI input/output moderators, Memory Manager=BufferMem2025.04 | 2 | 0.218 | 0.538 | — | |
| Chain-of-AttackAttack Protocol=Re-use attack, Safety Filter=OpenAI input/output moderators, Memory Manager=BufferMem2025.04 | 2.3 | 0.241 | 0.615 | — | |
| Chain-of-AttackAttack Protocol=One-time attack, Safety Filter=OpenAI input/output moderators, Memory Manager=BufferMem2025.04 | 5.3 | 0.247 | 0.617 | 23.6 | |
| MMADiffusionAttack Protocol=One-time attack, Safety Filter=OpenAI input/output moderators, Memory Manager=BufferMem2025.04 | 7.3 | 0.253 | 0.617 | — | |
| MMADiffusionAttack Protocol=Re-use attack, Safety Filter=OpenAI input/output moderators, Memory Manager=BufferMem2025.04 | 7.3 | 0.246 | 0.61 | — | |
| SneakyPromptAttack Protocol=Re-use attack, Safety Filter=OpenAI input/output moderators, Memory Manager=BufferMem2025.04 | 9.3 | 0.264 | 0.674 | — | |
| SneakyPromptAttack Protocol=One-time attack, Safety Filter=OpenAI input/output moderators, Memory Manager=BufferMem2025.04 | 10 | 0.258 | 0.668 | 34.19 | |
| DACAAttack Protocol=Re-use attack, Safety Filter=OpenAI input/output moderators, Memory Manager=BufferMem2025.04 | 12.7 | 0.249 | 0.624 | — | |
| DACAAttack Protocol=One-time attack, Safety Filter=OpenAI input/output moderators, Memory Manager=BufferMem2025.04 | 16.7 | 0.267 | 0.634 | — | |
| InceptionAttack Protocol=Re-use attack, Safety Filter=OpenAI input/output moderators, Memory Manager=BufferMem2025.04 | 21.3 | 0.268 | 0.679 | — | |
| InceptionAttack Protocol=One-time attack, Safety Filter=OpenAI input/output moderators, Memory Manager=BufferMem2025.04 | 28.7 | 0.278 | 0.701 | 10.26 |