ResearchBenchmarksPrompt Injection Attack on Real-world Overall (test)Follow64.8ASRPI3D60.43261.56662.763.834Feb 6, 2026Evaluation ResultsMethodMethodLinksASRPI3DEvaluator MLLM=GPT-4o-...Evaluator MLLM=GPT-4o-mini2026.0264.8PI3DEvaluator MLLM=Gemini-...Evaluator MLLM=Gemini-2.0-Flash2026.0260.6