Jailbreak Attack on Welsh Single-Turn 1.0 (test)
84Attack Success Rate (ASR)Claude 3 Sonnet
Evaluation Results
| Method | Links | |
|---|---|---|
| Claude 3 SonnetTurn Type=Single-Turn, Format=Welsh2025.04 | 84 | |
| Claude 3 HaikuTurn Type=Single-Turn, Format=Welsh2025.04 | 61 | |
| GPT-4 PreviewTurn Type=Single-Turn, Format=Welsh2025.04 | 58 | |
| GPT-3.5 TurboTurn Type=Single-Turn, Format=Welsh2025.04 | 48 | |
| Claude 3 OpusTurn Type=Single-Turn, Format=Welsh2025.04 | 11 |