ResearchBenchmarksAction Interception on DesktopTOCTOU-Bench Primitive AFollow100AIRPUSV99.27299.46199.6599.839Apr 20, 2026Evaluation ResultsMethodMethodLinksAIRPUSVVictim Agent Model=Cla...Victim Agent Model=Claude 4.6, Defense Status=ON, Trials per primitive=452026.04100PUSVVictim Agent Model=GPT...Victim Agent Model=GPT-4o, Defense Status=ON, Trials per primitive=452026.04100PUSVVictim Agent Model=Qwe...Victim Agent Model=Qwen3.6-plus, Defense Status=ON, Trials per primitive=452026.0499.3