Intent Shifting on Tool selection benchmark
80Certified Upper Bound (pu)SecAlign
Evaluation Results
| Method | Links | |
|---|---|---|
| SecAlignMitigation=SecAlign, Selector model=Llama-3.1, Generator model=Gemma-32025.10 | 80 | |
| StruQMitigation=StruQ, Selector model=Llama-3.1, Generator model=Gemma-32025.10 | 79 | |
| Anomaly mon.Mitigation=Anomaly mon., Selector model=Llama-3.1, Generator model=Gemma-32025.10 | 77 | |
| PerplexityMitigation=Perplexity, Selector model=Llama-3.1, Generator model=Gemma-32025.10 | 76 | |
| NoneMitigation=None, Selector model=Llama-3.1, Generator model=Gemma-32025.10 | 75 | |
| Defended retr.Mitigation=Defended retr., Selector model=Llama-3.1, Generator model=Gemma-32025.10 | 75 |