List your company

Adversarial Attack Detection on Balanced 500-sample non-adaptive

94PrecisionSpectralGuard (Multi-Layer)
4.5627.785174.22Mar 12, 2026

Evaluation results

MethodLinks
SpectralGuard (Multi-Layer)Setting=Non-adaptive, Attacker=Non-adaptive, Latency=< 15ms/tok, N=500, Model=Mamba-130M2026.03949896.10.9896
Internal L2-Norm MonitorSetting=Non-adaptive, Attacker=Non-adaptive, Latency=< 5ms, N=500, Model=Mamba-130M2026.03625156—5
SpectralGuard (Single-Layer)Setting=Non-adaptive, Attacker=Non-adaptive, Latency=< 5ms/tok, N=500, Model=Mamba-130M2026.0354.97061.9—67.2
Perplexity FilterSetting=Non-adaptive, Attacker=Non-adaptive, Latency=< 1ms, N=500, Model=Mamba-130M2026.03423136—2
Pattern MatcherSetting=Non-adaptive, Attacker=Non-adaptive, Latency=< 1ms, N=500, Model=Mamba-130M2026.03151213——
Toxicity ClassifierSetting=Non-adaptive, Attacker=Non-adaptive, Latency=< 1ms, N=500, Model=Mamba-130M2026.03856—1