ResearchBenchmarksSecure LLM Agent Task Completion on ASBFollow78.75Benign UtilityDRIFT23.3737.747552.12566.5025Jun 13, 2025Evaluation ResultsMethodMethodLinksBenign UtilityAttacked UtilityAttack Success Rate (ASR)DRIFTBase Model=GPT-4oBase Model=GPT-4o2025.0678.7569.758.5ProgentBase Model=GPT-4oBase Model=GPT-4o2025.067869.258DRIFTBase Model=GPT-4o-miniBase Model=GPT-4o-mini2025.0626.528.54.75ProgentBase Model=GPT-4o-miniBase Model=GPT-4o-mini2025.0625.528.515.75