Workflow Steering Defense on MisinfoTask
75.92TASRG-Safeguard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| G-SafeguardModel=Gemini-2.5-Flash2026.05 | 75.92 | 73.15 | |
| ARGUSModel=Gemini-2.5-Flash2026.05 | 64.81 | 73.15 | |
| w/o DefenseModel=GPT-4o-mini2026.05 | 63.89 | 65.74 | |
| w/o DefenseModel=Gemini-2.5-Flash2026.05 | 63.89 | 63.89 | |
| ARGUSModel=GPT-4o-mini2026.05 | 60.19 | 61.11 | |
| G-SafeguardModel=GPT-4o-mini2026.05 | 55.56 | 63.89 | |
| w/o DefenseModel=DeepSeek-V32026.05 | 51.85 | 56.48 | |
| ARGUSModel=Qwen-3.5-Flash2026.05 | 50 | 53.7 | |
| G-SafeguardModel=DeepSeek-V32026.05 | 48.15 | 53.7 | |
| G-SafeguardModel=Qwen-3.5-Flash2026.05 | 47.22 | 52.78 | |
| w/o DefenseModel=Qwen-3.5-Flash2026.05 | 46.29 | 51.85 | |
| ARGUSModel=DeepSeek-V32026.05 | 41.67 | 49.07 | |
| FLOWGUARDModel=Gemini-2.5-Flash2026.05 | 35.19 | 31.48 | |
| FLOWGUARDModel=GPT-4o-mini2026.05 | 33.33 | 35.19 | |
| FLOWGUARDModel=Qwen-3.5-Flash2026.05 | 27.78 | 38.89 | |
| FLOWGUARDModel=DeepSeek-V32026.05 | 19.44 | 22.22 |