Security Investigation Verdict Classification on Security Alert Logs Malicious
100Malicious RateAgentic security investigation workflow
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Agentic security investigation workflowModel=GPT-5-mini2026.04 | 100 | 0 | 0 | |
| Agentic security investigation workflowModel=Qwen3:30B2026.04 | 95 | 3 | 2 | |
| Agentic security investigation workflowModel=Claude 3 Haiku2026.04 | 94 | 0 | 6 | |
| Agentic security investigation workflowModel=Gemma 3:27B2026.04 | 81 | 4 | 15 |