Safety Compliance Evaluation on Mind2Web SC
95.8LPASpider-Sense
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Spider-SenseDefense Category=Guardrail-based, Base Agent=Claude-3.52026.02 | 95.8 | 88.7 | 100 | 92.1 | 100 | |
| AGrailDefense Category=Guardrail-based, Base Agent=Claude-3.52026.02 | 94 | 91.4 | 97 | 94.1 | 95.8 | |
| Spider-SenseDefense Category=Guardrail-based, Base Agent=Qwen-max2026.02 | 91.9 | 92.3 | 100 | 94.1 | 100 | |
| GuardAgentDefense Category=Guardrail-based, Base Agent=Claude-3.52026.02 | 85.7 | 86.8 | 78.7 | 74.3 | 91.3 | |
| Claude-3.5Defense Category=Model-based2026.02 | 84.8 | 84.3 | 100 | 90.3 | 99 | |
| GuardAgentDefense Category=Guardrail-based, Base Agent=Qwen-max2026.02 | 84.7 | 86.9 | 79.2 | 84.6 | 91.2 | |
| gpt-oss-safeguard-20bDefense Category=Guardrail-based2026.02 | 81.3 | 82.9 | 75 | 80 | 74.3 | |
| AGrailDefense Category=Guardrail-based, Base Agent=Qwen-max2026.02 | 81.3 | 72.7 | 100 | 84.2 | 96.3 | |
| Qwen-maxDefense Category=Model-based2026.02 | 80.1 | 85 | 78.1 | 82.6 | 76.2 | |
| LLaMA-Guard 3Defense Category=Guardrail-based2026.02 | 56 | 93 | 13 | 23 | — |