Safety Evaluation on Hazardous Viewpoints Dataset
10VSScoreRepackaged Statement
Evaluation Results
| Method | Links | |
|---|---|---|
| Repackaged StatementAttacker LLM=Qwen-2.5-7B, Category=Age2026.06 | 10 | |
| GhostwriterAttacker LLM=Qwen-2.5-7B, Target LLM=DeepSeek-V3, Query Type=Explicit Query, Category=Age2026.06 | 9.8 | |
| GhostwriterAttacker LLM=Qwen-2.5-7B, Target LLM=DeepSeek-V3, Query Type=Implicit Query, Category=Age2026.06 | 9.7 | |
| GhostwriterAttacker LLM=Qwen-2.5-7B, Target LLM=Claude-3.7-Sonnet, Query Type=Implicit Query, Category=Age2026.06 | 9.6 | |
| Repackaged StatementAttacker LLM=GPT-4o-mini, Category=Age2026.06 | 9.5 | |
| GhostwriterAttacker LLM=Qwen-2.5-7B, Target LLM=Claude-3.7-Sonnet, Query Type=Explicit Query, Category=Age2026.06 | 9.4 | |
| GhostwriterAttacker LLM=GPT-4o-mini, Target LLM=DeepSeek-V3, Query Type=Implicit Query, Category=Age2026.06 | 9 | |
| GhostwriterAttacker LLM=GPT-4o-mini, Target LLM=Claude-3.7-Sonnet, Query Type=Explicit Query, Category=Age2026.06 | 8.9 | |
| GhostwriterAttacker LLM=GPT-4o-mini, Target LLM=Claude-3.7-Sonnet, Query Type=Implicit Query, Category=Age2026.06 | 8.9 | |
| GhostwriterAttacker LLM=Qwen-2.5-7B, Target LLM=GPT-4o, Query Type=Explicit Query, Category=Age2026.06 | 8.9 | |
| GhostwriterAttacker LLM=GPT-4o-mini, Target LLM=DeepSeek-V3, Query Type=Explicit Query, Category=Age2026.06 | 8.7 | |
| Repackaged StatementAttacker LLM=GPT-4o-mini, Category=Gender2026.06 | 8.6 | |
| GhostwriterAttacker LLM=GPT-4o-mini, Target LLM=GPT-4o, Query Type=Explicit Query, Category=Age2026.06 | 8.4 | |
| GhostwriterAttacker LLM=Qwen-2.5-7B, Target LLM=GPT-4o, Query Type=Implicit Query, Category=Age2026.06 | 8.4 | |
| GhostwriterAttacker LLM=GPT-4o-mini, Target LLM=GPT-4o, Query Type=Implicit Query, Category=Age2026.06 | 8 | |
| GhostwriterAttacker LLM=Qwen-2.5-7B, Target LLM=DeepSeek-V3, Query Type=Implicit Query, Category=Psych. Manip.2026.06 | 8 | |
| GhostwriterAttacker LLM=GPT-4o-mini, Target LLM=GPT-4o, Query Type=Explicit Query, Category=Gender2026.06 | 6.5 | |
| No AttackTarget LLM=DeepSeek-V3, Query Type=Explicit Query, Category=Age2026.06 | 3 | |
| No AttackTarget LLM=DeepSeek-V3, Query Type=Implicit Query, Category=Age2026.06 | 2.6 | |
| No AttackTarget LLM=GPT-4o, Query Type=Implicit Query, Category=Age2026.06 | 2.5 | |
| No AttackTarget LLM=GPT-4o, Query Type=Explicit Query, Category=Age2026.06 | 2.2 | |
| No AttackTarget LLM=Claude-3.7-Sonnet, Query Type=Implicit Query, Category=Age2026.06 | 2.1 | |
| No AttackTarget LLM=Claude-3.7-Sonnet, Query Type=Explicit Query, Category=Age2026.06 | 1.8 |