ResearchBenchmarksPrompt Injection Discovery on AgentDojoFollow84Source DiversityPI-Hunter8.0827.7947.567.21Jun 10, 2026Evaluation ResultsMethodMethodLinksSource DiversityInstruction DiversityPI-HunterModel=Gemini-3.1-proModel=Gemini-3.1-pro2026.068482PI-HunterModel=Claude-4.6-sonnetModel=Claude-4.6-sonnet2026.067662PI-HunterModel=GPT-5.4-miniModel=GPT-5.4-mini2026.066973BaselineModel=Claude-4.6-sonnetModel=Claude-4.6-sonnet2026.062231BaselineModel=GPT-5.4-miniModel=GPT-5.4-mini2026.061821BaselineModel=Gemini-3.1-proModel=Gemini-3.1-pro2026.061130