siming

dahan6

Behavioral Grammar: detecting adaptive malware via tiny language-model priors and second-order temporal analysis. 0.88M-param self-supervised HIDS — 93% detection vs adaptive adversarial agent at 3.84% onboarding FPR.