List your company

Prompt Injection Detection on AgentDojo (Precision/Recall Metrics)

92.1Source PrecisionPI-Hunter
33.75648.90364.0579.197Jun 10, 2026

Evaluation results

MethodLinks
PI-HunterModel=Gemini-3.1-pro, Attack=important_inst2026.0692.189.689.481.3
PI-HunterModel=Gemini-3.1-pro, Attack=ignore_previous2026.069291.592.287.1
PI-HunterModel=Gemini-3.1-pro, Attack=system_msg2026.0690.789.490.887.1
PI-HunterModel=Claude-4.6-sonnet, Attack=ignore_previous2026.0689.181.693.877.8
PI-HunterModel=Claude-4.6-sonnet, Attack=direct2026.0687.771.495.782.5
PI-HunterModel=Claude-4.6-sonnet, Attack=system_msg2026.0687.770.992.869.3
PI-HunterModel=Gemini-3.1-pro, Attack=direct2026.0687.388.186.884
PI-HunterModel=Claude-4.6-sonnet, Attack=important_inst2026.068764.390.574.2
BaselineModel=Claude-4.6-sonnet, Attack=direct2026.0685.461.291.772.9
BaselineModel=Claude-4.6-sonnet, Attack=important_inst2026.0683.959.284.467.7
PI-HunterModel=GPT-5.4-mini, Attack=direct2026.0683.460.478.874.5
BaselineModel=Claude-4.6-sonnet, Attack=ignore_previous2026.0682.957.986.863.5
BaselineModel=Claude-4.6-sonnet, Attack=system_msg2026.0682.651.985.464.6
PI-HunterModel=GPT-5.4-mini, Attack=ignore_previous2026.0681.277.18282.6
PI-HunterModel=GPT-5.4-mini, Attack=agentvigil2026.0680.367.175.365
PI-HunterModel=Gemini-3.1-pro, Attack=agentvigil2026.0679.683.480.682.4
BaselineModel=GPT-5.4-mini, Attack=direct2026.0677.349.771.757.3
PI-HunterModel=Claude-4.6-sonnet, Attack=agentvigil2026.0677.368.180.757
PI-HunterModel=GPT-5.4-mini, Attack=system_msg2026.0675.467.68084.5
PI-HunterModel=GPT-5.4-mini, Attack=important_inst2026.0673.461.970.784.8
BaselineModel=Claude-4.6-sonnet, Attack=agentvigil2026.0673.451.672.549.6
BaselineModel=GPT-5.4-mini, Attack=ignore_previous2026.0671.948.273.160.4
BaselineModel=GPT-5.4-mini, Attack=agentvigil2026.0670.355.568.857.2
BaselineModel=GPT-5.4-mini, Attack=system_msg2026.0659.446.456.944.8
BaselineModel=GPT-5.4-mini, Attack=important_inst2026.065936.555.843.7
BaselineModel=Gemini-3.1-pro, Attack=system_msg2026.0657.626.772.442.7
BaselineModel=Gemini-3.1-pro, Attack=agentvigil2026.0655.825.567.543.6
BaselineModel=Gemini-3.1-pro, Attack=important_inst2026.0650.524.168.837.5
BaselineModel=Gemini-3.1-pro, Attack=direct2026.064924.860.942.2
BaselineModel=Gemini-3.1-pro, Attack=ignore_previous2026.063622.15034.8