List your company

Multi-turn Safety Risk Assessment on Terminal tasks

96ASRw/o Defense
12.834.45677.6Feb 13, 2026

Evaluation results

MethodLinks
w/o DefenseModel=Gemini-1.5-Flash2026.02960
FirewallModel=Gemini-1.5-Flash2026.02924
w/o DefenseModel=Claude-3.5-Sonnet2026.027624
FirewallModel=Claude-3.5-Sonnet2026.027228
BaselineModel=Gemini-1.5-Flash2026.025232
ToolShieldModel=Gemini-1.5-Flash2026.024444
BaselineModel=Claude-3.5-Sonnet2026.024052
ToolShieldModel=Claude-3.5-Sonnet2026.021684