Harmful Action Evaluation on Insider Threat
0Harmful Action RateInvThink SFT+RL
Evaluation Results
| Method | Links | |
|---|---|---|
| InvThink SFT+RLBackbone=Gemma-7B-it2025.10 | 0 | |
| InvThink SFT+RLBackbone=Qwen-2.5-7B2025.10 | 0 | |
| InvThink SFT+RLBackbone=Qwen-3-8B2025.10 | 0 | |
| General SFT+RLBackbone=Gemma-7B-it2025.10 | 1 | |
| InvThink SFTBackbone=Gemma-7B-it2025.10 | 1 | |
| InvThink SFTBackbone=Qwen-3-8B2025.10 | 1 | |
| General SFTBackbone=Gemma-7B-it2025.10 | 2 | |
| InvThinkBackbone=Qwen-2.5-7B2025.10 | 2 | |
| General SFT+RLBackbone=Qwen-2.5-7B2025.10 | 2 | |
| InvThink SFTBackbone=Qwen-2.5-7B2025.10 | 2 | |
| InvThinkBackbone=Qwen-3-8B2025.10 | 2 | |
| General SFT+RLBackbone=Qwen-3-8B2025.10 | 2 | |
| InvThinkBackbone=Gemma-7B-it2025.10 | 3 | |
| SafetyPromptBackbone=Qwen-2.5-7B2025.10 | 3 | |
| Gemini-2.5 ProBackbone=Teacher Model2025.10 | 3 | |
| SafetyPromptBackbone=Gemma-7B-it2025.10 | 4 | |
| Zero-shotBackbone=Qwen-2.5-7B2025.10 | 4 | |
| General SFTBackbone=Qwen-3-8B2025.10 | 4 | |
| CoTBackbone=Gemma-7B-it2025.10 | 5 | |
| CoTBackbone=Qwen-2.5-7B2025.10 | 5 | |
| General SFTBackbone=Qwen-2.5-7B2025.10 | 5 | |
| SafetyPromptBackbone=Qwen-3-8B2025.10 | 6 | |
| Zero-shotBackbone=Gemma-7B-it2025.10 | 7 | |
| ToTBackbone=Qwen-2.5-7B2025.10 | 7 | |
| Zero-shotBackbone=Qwen-3-8B2025.10 | 7 | |
| CoTBackbone=Qwen-3-8B2025.10 | 10 | |
| ToTBackbone=Qwen-3-8B2025.10 | 11 | |
| ToTBackbone=Gemma-7B-it2025.10 | 12 |