Safety Alignment Bypass on Platform-specific Safety Probes
55Bypass RateGhostwriter
Evaluation Results
| Method | Links | |
|---|---|---|
| GhostwriterCategory=Third-Party, Platform=Character.AI, Attack Point=System Prompt2026.06 | 55 | |
| GhostwriterCategory=Third-Party, Platform=Poe, Attack Point=System Prompt2026.06 | 100 | |
| GhostwriterCategory=Third-Party, Platform=Hugging Face Spaces, Attack Point=User Input2026.06 | 100 | |
| GhostwriterCategory=Third-Party, Platform=Cursor, Attack Point=MCP Server2026.06 | 100 | |
| GhostwriterCategory=Official, Platform=OpenAI GPT Store, Attack Point=System Prompt2026.06 | 100 | |
| GhostwriterCategory=Official, Platform=Claude Desktop, Attack Point=MCP Server2026.06 | 100 | |
| GhostwriterCategory=Social Media, Platform=Discord, Attack Point=User Input2026.06 | 100 | |
| GhostwriterCategory=Social Media, Platform=Telegram, Attack Point=User Input2026.06 | 100 | |
| GhostwriterCategory=Social Media, Platform=Twitter (X.com), Attack Point=User Input2026.06 | 100 |