Agentic Attack on WhisperBench Foreground execution (held-out)
98.2ISRMEMGHOST
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MEMGHOSTTarget Model=MiniMax-M2.5, Evaluation Cases=562026.07 | 98.2 | 21.4 | 85.7 | 21.4 | |
| MEMGHOSTTarget Model=DeepSeek v4 Flash, Evaluation Cases=562026.07 | 96.4 | 51.8 | 85.7 | 42.9 | |
| MEMGHOSTTarget Model=GPT-5.4, Evaluation Cases=562026.07 | 94.6 | 82.1 | 87.5 | 75 | |
| MEMGHOSTTarget Model=Sonnet 4.5, Evaluation Cases=562026.07 | 92.9 | 80.4 | 89.3 | 71.4 | |
| MEMGHOSTTarget Model=Grok 4.1 Fast, Evaluation Cases=562026.07 | 89.3 | 96.4 | 80.4 | 76.8 | |
| MEMGHOSTTarget Model=Sonnet 4.6, Evaluation Cases=562026.07 | 51.8 | 51.8 | 48.2 | 48.2 | |
| RGSTarget Model=GPT-5.4, Evaluation Cases=202026.07 | 35 | 25 | 35 | 20 | |
| Naive AttackTarget Model=GPT-5.4, Evaluation Cases=562026.07 | 0 | 0 | 0 | 0 |