Adversarial Hallucination Attack on MMLU filtered subset (16 subjects)
2.01ASR@30Raw prompting
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Raw promptingTarget Model=GPT-5-Mini2026.05 | 2.01 | 1.58 | 0 | |
| ICDTarget Model=GPT-5-Mini2026.05 | 2.57 | 13.97 | 100 | |
| Raw promptingTarget Model=GPT-5-Nano2026.05 | 4.02 | 1.58 | 0 | |
| ICDTarget Model=GPT-5-Nano2026.05 | 6.32 | 13.97 | 100 | |
| REALISTATarget Model=GPT-5-Mini2026.05 | 20.72 | 0.72 | 0.87 | |
| REALISTATarget Model=GPT-5-Nano2026.05 | 23.61 | 1.59 | 1.73 |