Jailbreak Detection on GoalFrameBench harmful Llama3-8B 2025 (seed prompts)
0.96AccuracyFrameShield-Last
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FrameShield-LastBackbone=Llama3-8B2026.02 | 0.96 | 0.86 | |
| JBShieldBackbone=Llama3-8B2026.02 | 0.77 | 0.86 | |
| LlamaGuardBackbone=Llama3-8B2026.02 | 0.6 | 0.75 | |
| GradSafeBackbone=Llama3-8B2026.02 | 0.37 | 0.54 | |
| SelfExBackbone=Llama3-8B2026.02 | 0.16 | 0.26 |