Attack Detection on Jailbreak Attacks 105K sample set
68Detection RateLogReg (Ours)
Evaluation Results
| Method | Links | |
|---|---|---|
| LogReg (Ours)input=raw activations, evaluation_regime=LODO2026.02 | 68 | |
| Llama-as-Judgeacronym=LJ, prompting=zero-shot2026.02 | 60 | |
| PromptGuard 2acronym=PG2026.02 | 48.5 | |
| LlamaGuardacronym=LG2026.02 | 28.9 |