List your company

Tool Attack Defense on InjecAgent random topology (test)

0.375ASR@1No Defense
0.008920.103960.1990.29404Jan 21, 2026

Evaluation results

MethodLinks
No DefenseBackbone=GPT-4o-mini2026.010.3750.5580.6750.6330.5470.333
AgentXposed-KickBackbone=GPT-4o-mini2026.010.340.480.5030.6830.5170.517
AgentXposed-GuideBackbone=GPT-4o-mini2026.010.3230.4330.490.70.5670.533
No DefenseBackbone=Qwen3-235B-A22B2026.010.1770.150.1430.8670.8670.867
InspectorBackbone=GPT-4o-mini2026.010.150.2130.2430.8670.7830.75
ChallengerBackbone=GPT-4o-mini2026.010.1130.2130.2770.90.80.717
AgentXposed-KickBackbone=Qwen3-235B-A22B2026.010.1030.0870.0770.9330.9330.933
G-safeguardBackbone=GPT-4o-mini2026.010.0930.1260.1310.90.8870.887
InspectorBackbone=Qwen3-235B-A22B2026.010.090.0630.0470.950.950.967
G-safeguardBackbone=Qwen3-235B-A22B2026.010.0730.0670.0730.9770.9330.933
AgentXposed-GuideBackbone=Qwen3-235B-A22B2026.010.0730.0430.050.950.9670.967
AgentSafeBackbone=GPT-4o-mini2026.010.0630.0770.120.950.9330.883
INFA-GUARDBackbone=GPT-4o-mini2026.010.0620.0310.0210.9130.9830.983
ChallengerBackbone=Qwen3-235B-A22B2026.010.050.050.070.9830.950.933
INFA-GUARDBackbone=Qwen3-235B-A22B2026.010.0430.0230.0370.9830.9830.983
AgentSafeBackbone=Qwen3-235B-A22B2026.010.0230.010.0030.98311