ResearchBenchmarksTargeted Failure Generation on ImageNetFor AI assistantsMarkdownContent indexTargeted Failure Generation on ImageNetFollow0.82Misclassification RateMimicry0.81280.86140.910.9586Jan 21, 2026Evaluation resultsMethodMethodLinksMisclassification RateEscape RatioConfidence ReductionMS-SSIMLPIPSDiversity Score (T)Trace DifferenceMimicry2026.010.820.52———0.147197.21HyNeA2026.0110———0.17979.6GIFTbench2026.0110.84———0.102119.92