List your company

Prompt Injection Defense on AgentDojo New Attack 2

89.78Utility under Attack (UA)No Defense
7.807229.088650.3771.6514Feb 8, 2026

Evaluation results

MethodLinks
No DefenseBackbone Model=Gemini-3-Pro2026.0289.7818.85
Repeat PromptBackbone Model=Gemini-3-Pro2026.0289.361.216.11
CausalArmorBackbone Model=Gemini-3-Pro2026.0287.881.530
PromptArmorBackbone Model=Gemini-3-Pro2026.0283.772.050.32
CausalArmorBackbone Model=Claude-4-Sonnet2026.0283.031.70.46
No DefenseBackbone Model=Claude-4-Sonnet2026.0281.77118.97
PromptArmorBackbone Model=Claude-4-Sonnet2026.0275.872.660
Repeat PromptBackbone Model=Claude-4-Sonnet2026.0267.331.247.27
DRIFTBackbone Model=Gemini-3-Pro2026.0266.686.534.24
DeBERTa DetectorBackbone Model=Gemini-3-Pro2026.0256.81.066.11
DRIFTBackbone Model=Claude-4-Sonnet2026.0255.537.592.48
DeBERTa DetectorBackbone Model=Claude-4-Sonnet2026.0255.321.146.53
CausalArmorBackbone Model=Gemini-2.5-Flash2026.0252.531.810.4
PromptArmorBackbone Model=Gemini-2.5-Flash2026.0251.742.710.24
Repeat PromptBackbone Model=Gemini-2.5-Flash2026.0251.531.147.38
No DefenseBackbone Model=Gemini-2.5-Flash2026.0247.4817.06
MelonBackbone Model=Claude-4-Sonnet2026.0242.472.720
DRIFTBackbone Model=Gemini-2.5-Flash2026.0242.445.251.25
MelonBackbone Model=Gemini-2.5-Flash2026.0230.982.671.05
PiGuardBackbone Model=Gemini-3-Pro2026.0229.921.070.84
DeBERTa DetectorBackbone Model=Gemini-2.5-Flash2026.0228.351.244.11
PiGuardBackbone Model=Claude-4-Sonnet2026.0227.921.20.42
PiGuardBackbone Model=Gemini-2.5-Flash2026.0210.961.30