List your company

Adversarial and Jailbreaking Attack Detection on XSTest

0.984AUROCGradSafe
0.12340.3468250.570250.793675Feb 4, 2026Feb 18, 2026Mar 5, 2026Mar 19, 2026Apr 3, 2026Apr 17, 2026May 2, 2026

Evaluation results

MethodLinks
GradSafeTarget Model=Llama-3.1-8B-Instruct, Grad=true2026.050.984—
Linear ProbeTarget Model=Mistral-7B-Instruct-v0.3, Params=<1M, Grad=false2026.050.97—
GradSafeTarget Model=Mistral-7B-Instruct-v0.3, Grad=true2026.050.95—
Linear ProbeTarget Model=Llama-3.1-8B-Instruct, Params=<1M, Grad=false2026.050.949—
SALOTarget Model=Llama-3.1-8B-Instruct, Params=<20M, Grad=false2026.050.948—
SALOTarget Model=Qwen2.5-7B-Instruct, Params=<20M, Grad=false2026.050.946—
GradSafeTarget Model=Qwen2.5-7B-Instruct, Grad=true2026.050.942—
SALOTarget Model=Mistral-7B-Instruct-v0.3, Params=<20M, Grad=false2026.050.932—
Linear ProbeTarget Model=Qwen2.5-7B-Instruct, Params=<1M, Grad=false2026.050.93—
Smooth LLMTarget Model=Qwen2.5-7B-Instruct, Grad=false2026.050.893—
DUOGUARDMode=Zero-shot2026.020.84180.781
OpenAI OmniMode=Zero-shot2026.020.82570.9667
LLAMAGUARD4-12BMode=Zero-shot2026.020.8121
FDBDMode=Zero-shot2026.020.80210.781
POLYGUARDMode=Zero-shot2026.020.80070.8714
Perspective APIMode=Zero-shot2026.020.79320.7381
WILDGUARDMode=Zero-shot2026.020.79291
MDJUDGEMode=Zero-shot2026.020.79060.9238
Smooth LLMTarget Model=Mistral-7B-Instruct-v0.3, Grad=false2026.050.786—
LLAMAGUARD3-1BMode=Zero-shot2026.020.78431
Smooth LLMTarget Model=Llama-3.1-8B-Instruct, Grad=false2026.050.716—
T3-GMMMode=Zero-shot2026.020.67940.8143
LLAMAGUARD3-1B-LOGITSMode=Zero-shot2026.020.65420.9095
OPENMAXMode=Zero-shot2026.020.63080.9952
PPL FilterTarget Model=Qwen2.5-7B-Instruct, Grad=false2026.050.595—
T3-OCSVMMode=Zero-shot2026.020.580.8762
PPL FilterTarget Model=Mistral-7B-Instruct-v0.3, Grad=false2026.050.57—
PPL FilterTarget Model=Llama-3.1-8B-Instruct, Grad=false2026.050.481—
RMDMode=Zero-shot2026.020.3731
CIDERMode=Zero-shot2026.020.33451
ADASCALEMode=Zero-shot2026.020.28030.9952
REACTMode=Zero-shot2026.020.23731
GMMMode=Zero-shot2026.020.22981
NNGUIDEMode=Zero-shot2026.020.21441
VIMMode=Zero-shot2026.020.15651