Tool-use behavioral profiling on Malicious regime n=30
50Metric Amistral:latest
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| mistral:latestAutonomy=A02026.04 | 50 | 40 | 10 | |
| mistral:latestAutonomy=A12026.04 | 50 | 27 | 23 | |
| deepseek-chatAutonomy=A02026.04 | 43 | 57 | 0 | |
| mistral:latestAutonomy=A22026.04 | 43 | 23 | 33 | |
| qwen2.5:7b-instructAutonomy=A12026.04 | 30 | 63 | 20 | |
| qwen2.5:7b-instructAutonomy=A02026.04 | 20 | 60 | 20 | |
| deepseek-chatAutonomy=A12026.04 | 10 | 90 | 0 | |
| llama3.1:latestAutonomy=A12026.04 | 7 | 90 | 17 | |
| llama3.1:latestAutonomy=A22026.04 | 7 | 93 | 7 | |
| qwen2.5:7b-instructAutonomy=A22026.04 | 7 | 97 | 3 | |
| llama3.1:latestAutonomy=A02026.04 | 3 | 100 | 3 | |
| deepseek-chatAutonomy=A22026.04 | 0 | 100 | 0 |