Reasoning-Aware Threat Detection on Reasoning-Aware Security Evaluation Latent Triggers
0.992Det-F1GPT-5.2 + Instructions
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT-5.2 + InstructionsModel Class=Upper Bound, Detector Variant=GPT-5.2 + Instructions [28]2026.03 | 0.992 | 0.997 | |
| GPT-OSS-20B-GuardModel Class=TRACEGUARD (Ours), Detector Variant=GPT-OSS-20B-Guard2026.03 | 0.949 | 0.976 | |
| Qwen3-4B-GuardModel Class=TRACEGUARD (Ours), Detector Variant=Qwen3-4B-Guard2026.03 | 0.897 | 0.95 | |
| GPT-OSS-20BModel Class=CoS [22], Detector Variant=GPT-OSS-20B [29]2026.03 | 0.849 | 0.854 | |
| Qwen3-4BModel Class=CoS [22], Detector Variant=Qwen3-4B [14]2026.03 | 0.774 | 0.737 | |
| DeepSeek-R1-7B-GuardModel Class=TRACEGUARD (Ours), Detector Variant=DeepSeek-R1-7B-Guard2026.03 | 0.731 | 0.8 | |
| GPT-OSS-20BModel Class=Zero-Shot Baselines, Detector Variant=GPT-OSS-20B [29]2026.03 | 0.635 | 0.658 | |
| Qwen3-4BModel Class=Zero-Shot Baselines, Detector Variant=Qwen3-4B [14]2026.03 | 0.569 | 0.642 | |
| DeepSeek-R1-7BModel Class=CoS [22], Detector Variant=DeepSeek-R1-7B [3]2026.03 | 0.539 | 0.3 | |
| DeepSeek-R1-7BModel Class=Zero-Shot Baselines, Detector Variant=DeepSeek-R1-7B [3]2026.03 | 0.18 | 0.518 | |
| Llama-Guard-3-8BModel Class=Llama-Guard [26], Detector Variant=Llama-Guard-3-8B [26]2026.03 | 0 | — |