Perturbed Policy Refusal on General Data Protection Regulation (GDPR)
0Refusal RateGPT-5-MINI
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5-MINIPerturbation=Deontic Norm, DEF Stage=Correct2026.01 | 0 | |
| GPT-5-MINIPerturbation=Authorization, DEF Stage=Correct2026.01 | 0 | |
| GPT-5-MINIPerturbation=Authorization, DEF Stage=No DEF2026.01 | 0.5 | |
| GPT-5-MINIPerturbation=Authorization, DEF Stage=DEF 12026.01 | 0.5 | |
| GPT-5-MINIPerturbation=Authorization, DEF Stage=DEF 22026.01 | 1.1 | |
| GPT-5-MINIPerturbation=Deontic Norm, DEF Stage=DEF 12026.01 | 1.7 | |
| QWEN3-30B-THINKPerturbation=Authorization, DEF Stage=No DEF2026.01 | 2.1 | |
| QWEN3-30B-THINKPerturbation=Authorization, DEF Stage=DEF 22026.01 | 2.1 | |
| DEEPSEEK-R1Perturbation=Authorization, DEF Stage=No DEF2026.01 | 2.1 | |
| QWEN3-30B-THINKPerturbation=Deontic Norm, DEF Stage=Correct2026.01 | 2.3 | |
| QWEN3-30B-THINKPerturbation=Authorization, DEF Stage=Correct2026.01 | 2.3 | |
| DEEPSEEK-R1Perturbation=Deontic Norm, DEF Stage=Correct2026.01 | 2.3 | |
| DEEPSEEK-R1Perturbation=Authorization, DEF Stage=Correct2026.01 | 2.3 | |
| GPT-5-MINIPerturbation=Deontic Norm, DEF Stage=No DEF2026.01 | 2.5 | |
| DEEPSEEK-R1Perturbation=Authorization, DEF Stage=DEF 12026.01 | 2.9 | |
| QWEN3-30B-THINKPerturbation=Authorization, DEF Stage=DEF 12026.01 | 3.2 | |
| DEEPSEEK-R1Perturbation=Deontic Norm, DEF Stage=No DEF2026.01 | 3.4 | |
| DEEPSEEK-R1Perturbation=Deontic Norm, DEF Stage=DEF 12026.01 | 3.7 | |
| QWEN3-30B-THINKPerturbation=Authorization, DEF Stage=DEF 32026.01 | 4.3 | |
| QWEN3-30B-THINKPerturbation=Deontic Norm, DEF Stage=DEF 12026.01 | 6.8 | |
| QWEN3-30B-THINKPerturbation=Deontic Norm, DEF Stage=No DEF2026.01 | 7.6 | |
| DEEPSEEK-R1Perturbation=Authorization, DEF Stage=DEF 22026.01 | 11.2 | |
| QWEN3-30B-THINKPerturbation=Deontic Norm, DEF Stage=DEF 22026.01 | 16.9 | |
| QWEN3-30B-THINKPerturbation=Deontic Norm, DEF Stage=DEF 32026.01 | 17 | |
| GPT-5-MINIPerturbation=Authorization, DEF Stage=DEF 32026.01 | 23.4 | |
| DEEPSEEK-R1Perturbation=Deontic Norm, DEF Stage=DEF 22026.01 | 28.8 | |
| GPT-5-MINIPerturbation=Deontic Norm, DEF Stage=DEF 22026.01 | 35.6 | |
| DEEPSEEK-R1Perturbation=Authorization, DEF Stage=DEF 32026.01 | 59 | |
| DEEPSEEK-R1Perturbation=Deontic Norm, DEF Stage=DEF 32026.01 | 83 | |
| GPT-5-MINIPerturbation=Deontic Norm, DEF Stage=DEF 32026.01 | 92.4 |