List your company

Attack Detection on Benign Samples 105K sample set

40Benign FPRPromptGuard 2
15.6180.3345509.7Feb 15, 2026

Evaluation results

MethodLinks
PromptGuard 2acronym=PG2026.0240
LlamaGuardacronym=LG2026.02300
Llama-as-Judgeacronym=LJ, prompting=zero-shot2026.02440
LogReg (Ours)input=raw activations, evaluation_regime=LODO2026.02650