List your company

Adversarial and Jailbreaking Attack Detection on HarmBench

0.8887AUROCLLAMAGUARD3-1B-LOGITS
0.3148280.4638140.61280.761786Feb 4, 2026

Evaluation results

MethodLinks
LLAMAGUARD3-1B-LOGITSMode=Zero-shot2026.020.88870.36
LLAMAGUARD4-12BMode=Zero-shot2026.020.88681
LLAMAGUARD3-1BMode=Zero-shot2026.020.88571
WILDGUARDMode=Zero-shot2026.020.86421
POLYGUARDMode=Zero-shot2026.020.85950.645
OpenAI OmniMode=Zero-shot2026.020.81850.965
T3-OCSVMMode=Zero-shot2026.020.81020.585
DUOGUARDMode=Zero-shot2026.020.80070.955
MDJUDGEMode=Zero-shot2026.020.7980.905
T3-GMMMode=Zero-shot2026.020.75780.67
Perspective APIMode=Zero-shot2026.020.72470.95
FDBDMode=Zero-shot2026.020.63420.92
OPENMAXMode=Zero-shot2026.020.49040.99
RMDMode=Zero-shot2026.020.48461
CIDERMode=Zero-shot2026.020.47990.965
ADASCALEMode=Zero-shot2026.020.43410.99
GMMMode=Zero-shot2026.020.41630.99
NNGUIDEMode=Zero-shot2026.020.40230.985
REACTMode=Zero-shot2026.020.36481
VIMMode=Zero-shot2026.020.33691