List your company

Adversarial Hallucination Attack on MMLU filtered subset (16 subjects)

2.01ASR@30Raw prompting
1.1466.97812.8118.642May 12, 2026

Evaluation results

MethodLinks
Raw promptingTarget Model=GPT-5-Mini2026.052.011.580
ICDTarget Model=GPT-5-Mini2026.052.5713.97100
Raw promptingTarget Model=GPT-5-Nano2026.054.021.580
ICDTarget Model=GPT-5-Nano2026.056.3213.97100
REALISTATarget Model=GPT-5-Mini2026.0520.720.720.87
REALISTATarget Model=GPT-5-Nano2026.0523.611.591.73