List your company

Attack Detection on InjecAgent Agentic Attacks

0.99Detection RateLogReg (Ours)
0.1840.393250.60250.81175Feb 15, 2026

Evaluation results

MethodLinks
LogReg (Ours)input=raw activations, evaluation_regime=LODO2026.020.99
Llama-as-Judgeacronym=LJ, prompting=zero-shot2026.020.215