List your company

Attack Detection on Harmful Attacks 105K sample set

97.4Detection RateLlamaGuard
3.48827.86952.2576.631Feb 15, 2026

Evaluation results

MethodLinks
LlamaGuardacronym=LG2026.0297.4
Llama-as-Judgeacronym=LJ, prompting=zero-shot2026.0285.8
LogReg (Ours)input=raw activations, evaluation_regime=LODO2026.0268
LogReg (Ours)input=raw activations, evaluation_regime=LODO2026.0267
PromptGuard 2acronym=PG2026.0237.3
PromptGuard 2acronym=PG2026.0236.7
LlamaGuardacronym=LG2026.0227.4
Llama-as-Judgeacronym=LJ, prompting=zero-shot2026.027.1