List your company

Attack Detection on Extraction Attacks 105K sample set

100Detection RatePromptGuard 2
11.80834.70457.680.496Feb 15, 2026

Evaluation results

MethodLinks
PromptGuard 2acronym=PG2026.02100
LogReg (Ours)input=raw activations, evaluation_regime=LODO2026.0279
Llama-as-Judgeacronym=LJ, prompting=zero-shot2026.0231.8
LlamaGuardacronym=LG2026.0215.2