Phrase Protection on Instruction Hierarchy (test)
97.5User Message Protection AccuracyOpenAI o3
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OpenAI o32025.12 | 97.5 | 92.1 | |
| gpt-5-thinking2025.12 | 94 | 91.1 | |
| GPT-4o2025.12 | 73.5 | 44.9 | |
| gpt-5-main2025.12 | 61.9 | 40.4 |
| Method | Links | ||
|---|---|---|---|
| OpenAI o32025.12 | 97.5 | 92.1 | |
| gpt-5-thinking2025.12 | 94 | 91.1 | |
| GPT-4o2025.12 | 73.5 | 44.9 | |
| gpt-5-main2025.12 | 61.9 | 40.4 |