Embodied AI Jailbreaking on Blindfold
88.7ASRSafeDecoding
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SafeDecoding2026.03 | 88.7 | -4.8 | |
| Llama-Guard2026.03 | 86.1 | -7.6 | |
| VeriSafe2026.03 | 76.5 | -17.9 |
| Method | Links | ||
|---|---|---|---|
| SafeDecoding2026.03 | 88.7 | -4.8 | |
| Llama-Guard2026.03 | 86.1 | -7.6 | |
| VeriSafe2026.03 | 76.5 | -17.9 |