Deceptive Defense on EMRA (test)
0.387MTA (Average)CoopGuard
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CoopGuardBackbone=DeepSeek-V32026.04 | 0.387 | 0.342 | 0.546 | 0.273 | |
| CoopGuardBackbone=Gemini-2.5-Pro2026.04 | 0.334 | 0.323 | 0.454 | 0.224 | |
| CoopGuardBackbone=GPT-52026.04 | 0.303 | 0.137 | 0.485 | 0.287 | |
| Self-ReminderBackbone=DeepSeek-V32026.04 | 0.068 | 0.001 | 0.204 | 0 | |
| RPOBackbone=Gemini-2.5-Pro2026.04 | 0.057 | 0 | 0.168 | 0.004 | |
| PATBackbone=Gemini-2.5-Pro2026.04 | 0.044 | 0.01 | 0.122 | 0 | |
| RPOBackbone=DeepSeek-V32026.04 | 0.043 | 0.001 | 0.124 | 0.003 | |
| SecurityLinguaBackbone=Gemini-2.5-Pro2026.04 | 0.042 | 0.01 | 0.103 | 0.013 | |
| PATBackbone=DeepSeek-V32026.04 | 0.039 | 0.001 | 0.117 | 0 | |
| SecurityLinguaBackbone=GPT-52026.04 | 0.036 | 0.007 | 0.1 | 0 | |
| Self-ReminderBackbone=Gemini-2.5-Pro2026.04 | 0.025 | 0 | 0.076 | 0 | |
| Self-ReminderBackbone=GPT-52026.04 | 0.022 | 0 | 0.065 | 0 | |
| PATBackbone=GPT-52026.04 | 0.015 | 0.013 | 0.031 | 0 | |
| RPOBackbone=GPT-52026.04 | 0.014 | 0.005 | 0.018 | 0.019 | |
| GoalPriorityBackbone=DeepSeek-V32026.04 | 0.011 | 0.01 | 0.007 | 0.015 | |
| SecurityLinguaBackbone=DeepSeek-V32026.04 | 0.008 | 0.01 | 0.003 | 0.01 | |
| GoalPriorityBackbone=GPT-52026.04 | 0.003 | 0 | 0.009 | 0 | |
| GoalPriorityBackbone=Gemini-2.5-Pro2026.04 | 0.002 | 0 | 0.005 | 0 |