Jailbreak Defense Evaluation on EMRA JQ
1Attack Success Rate (ASR)GoalPriority
Evaluation Results
| Method | Links | |
|---|---|---|
| GoalPriorityTarget LLM=GPT-52026.04 | 1 | |
| CoopGuardTarget LLM=GPT-52026.04 | 1 | |
| GoalPriorityTarget LLM=Gemini-2.5-Pro2026.04 | 1 | |
| Self-ReminderTarget LLM=GPT-52026.04 | 1.8 | |
| RPOTarget LLM=GPT-52026.04 | 2.3 | |
| CoopGuardTarget LLM=Gemini-2.5-Pro2026.04 | 2.3 | |
| PATTarget LLM=GPT-52026.04 | 2.7 | |
| CoopGuardTarget LLM=DeepSeek-V32026.04 | 4.2 | |
| Self-ReminderTarget LLM=Gemini-2.5-Pro2026.04 | 5.1 | |
| GoalPriorityTarget LLM=DeepSeek-V32026.04 | 5.3 | |
| Self-ReminderTarget LLM=DeepSeek-V32026.04 | 12.6 | |
| SecurityLinguaTarget LLM=DeepSeek-V32026.04 | 17.4 | |
| SecurityLinguaTarget LLM=Gemini-2.5-Pro2026.04 | 21 | |
| RPOTarget LLM=Gemini-2.5-Pro2026.04 | 21.1 | |
| PATTarget LLM=Gemini-2.5-Pro2026.04 | 26.4 | |
| SecurityLinguaTarget LLM=GPT-52026.04 | 28.3 | |
| PATTarget LLM=DeepSeek-V32026.04 | 28.3 | |
| RPOTarget LLM=DeepSeek-V32026.04 | 31.2 |