Jailbreak Attack on Opposite-day (test)
10.5ASRGPT-5
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5Safety Steering=Enabled2025.02 | 10.5 | |
| Claude Sonnet 4.5Safety Steering=Enabled2025.02 | 13.5 | |
| Claude Sonnet 4.5Safety Steering=Disabled2025.02 | 25 | |
| GPT-5Safety Steering=Disabled2025.02 | 32.5 |