Jailbreak Attack on HarmBench standard behaviors
61ASRAdvGRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| AdvGRPOTransfer Defender=Llama-3.1-8B, Attacker=Qwen2.5-14B, Turn Type=Single-turn (ST)2026.06 | 61 | |
| AdvGRPOTransfer Defender=Phi-4-mini, Attacker=Qwen2.5-14B, Turn Type=Single-turn (ST)2026.06 | 67.5 | |
| AdvGRPOTransfer Defender=Gemma-2-9B, Attacker=Qwen2.5-14B, Turn Type=Single-turn (ST)2026.06 | 68.5 | |
| SEMATransfer Defender=Phi-4-mini, Attacker=Qwen2.5-14B2026.06 | 71.5 | |
| SEMATransfer Defender=Gemma-2-9B, Attacker=Qwen2.5-14B2026.06 | 75.5 | |
| SEMATransfer Defender=Llama-3.1-8B, Attacker=Qwen2.5-14B2026.06 | 79.5 | |
| AdvGRPOTransfer Defender=Phi-4-mini, Attacker=Qwen2.5-14B, Turn Type=Multi-turn (MT)2026.06 | 82 | |
| AdvGRPOTransfer Defender=Gemma-2-9B, Attacker=Qwen2.5-14B, Turn Type=Multi-turn (MT)2026.06 | 86 | |
| AdvGRPOTransfer Defender=Llama-3.1-8B, Attacker=Qwen2.5-14B, Turn Type=Multi-turn (MT)2026.06 | 88.5 |