Jailbreak Defense on Emoji Attack GCG
1Attack Success Rate (ASR)SelfDefend
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SelfDefendvariant=Direct, Target LLM=LLama-3-8B Instruct2026.04 | 1 | 8 | |
| SelfGraderTarget LLM=LLama-3-8B Instruct2026.04 | 2 | 4 | |
| Llama Guardvariant=Pre, Target LLM=LLama-3-8B Instruct2026.04 | 5 | 87 |