Data Steal on Book
100ASRTIP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TIPDefense=No Defense2026.03 | 100 | 20 | |
| TIPDefense=Finetuned Detector2026.03 | 97.8 | 20 | |
| TIPDefense=Perplexity Filtering2026.03 | 94 | 320 | |
| TIPDefense=Instruction Prevention2026.03 | 61 | 820 | |
| TIPDefense=Sandwich Prevention2026.03 | 36 | 2,420 | |
| TAPDefense=Finetuned Detector2026.03 | 20 | 2,580 | |
| FixedDefense=No Defense2026.03 | 10.2 | — | |
| FixedDefense=Sandwich Prevention2026.03 | 8.8 | — | |
| FixedDefense=Perplexity Filtering2026.03 | 7.5 | — | |
| TAPDefense=No Defense2026.03 | 0 | 2,580 | |
| FixedDefense=Instruction Prevention2026.03 | 0 | — | |
| TAPDefense=Instruction Prevention2026.03 | 0 | 2,580 | |
| TAPDefense=Sandwich Prevention2026.03 | 0 | 2,580 | |
| FixedDefense=Finetuned Detector2026.03 | 0 | — | |
| TAPDefense=Perplexity Filtering2026.03 | 0 | 2,580 |