Safety Detection on TS-Bench AgentHarm-Traj (eval)
1.36Latency (s/sample)TS-Guard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TS-Guard2026.01 | 1.36 | 90.16 | |
| GPT-4o2026.01 | 1.98 | 84.8 | |
| AGrailModel=GPT-4o2026.01 | 8.75 | 85.75 | |
| ShieldAgent2026.01 | 10 | — |
| Method | Links | ||
|---|---|---|---|
| TS-Guard2026.01 | 1.36 | 90.16 | |
| GPT-4o2026.01 | 1.98 | 84.8 | |
| AGrailModel=GPT-4o2026.01 | 8.75 | 85.75 | |
| ShieldAgent2026.01 | 10 | — |