Jailbreak Attack on Welsh Multi-Turn 1.0 (test)
72Attack Success Rate (ASR)Claude 3 Sonnet
Evaluation Results
| Method | Links | |
|---|---|---|
| Claude 3 SonnetTurn Type=Multi-Turn, Format=Welsh2025.04 | 72 | |
| Claude 3 HaikuTurn Type=Multi-Turn, Format=Welsh2025.04 | 30 | |
| GPT-3.5 TurboTurn Type=Multi-Turn, Format=Welsh2025.04 | 14 | |
| Claude 3 OpusTurn Type=Multi-Turn, Format=Welsh2025.04 | 8 | |
| GPT-4 PreviewTurn Type=Multi-Turn, Format=Welsh2025.04 | 4 |