Jailbreak Detection on GoalFrameBench harmful Llama2-7B 2025 (seed prompts)
95AccuracyFrameShield-Last
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FrameShield-LastBackbone=Llama2-7B2026.02 | 95 | 85 | |
| JBShieldBackbone=Llama2-7B2026.02 | 84 | 86 | |
| GradSafeBackbone=Llama2-7B2026.02 | 62 | 77 | |
| LlamaGuardBackbone=Llama2-7B2026.02 | 53 | 69 | |
| SelfExBackbone=Llama2-7B2026.02 | 27 | 30 |