Agent Safety Judgment on ROME IR unsafe
95.74F1 ScoreClaude Sonnet 4.6
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Claude Sonnet 4.6Model=Claude Sonnet 4.62026.05 | 95.74 | 91.84 | 98 | |
| GPT-5.2Model=GPT-5.22026.05 | 91.3 | 84 | 100 | |
| Claude Opus 4.6Model=Claude Opus 4.62026.05 | 86.06 | 75.53 | 94 | |
| GPT-5Model=GPT-52026.05 | 83.04 | 71 | 100 |