Reasoning-Aware Threat Detection on Reasoning-Aware Security Evaluation Fallacy Injections
100Det-F1GPT-5.2 + Instructions
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT-5.2 + InstructionsModel Class=Upper Bound, Detector Variant=GPT-5.2 + Instructions [28]2026.03 | 100 | 100 | |
| GPT-OSS-20B-GuardModel Class=TRACEGUARD (Ours), Detector Variant=GPT-OSS-20B-Guard2026.03 | 99.8 | 99.8 | |
| Qwen3-4B-GuardModel Class=TRACEGUARD (Ours), Detector Variant=Qwen3-4B-Guard2026.03 | 99.7 | 99.7 | |
| GPT-OSS-20BModel Class=CoS [22], Detector Variant=GPT-OSS-20B [29]2026.03 | 94.9 | 87.4 | |
| Qwen3-4BModel Class=CoS [22], Detector Variant=Qwen3-4B [14]2026.03 | 92.4 | 74.5 | |
| DeepSeek-R1-7BModel Class=CoS [22], Detector Variant=DeepSeek-R1-7B [3]2026.03 | 81.7 | 54.4 | |
| GPT-OSS-20BModel Class=Zero-Shot Baselines, Detector Variant=GPT-OSS-20B [29]2026.03 | 79.9 | 71.4 | |
| Qwen3-4BModel Class=Zero-Shot Baselines, Detector Variant=Qwen3-4B [14]2026.03 | 73.5 | 78.7 | |
| DeepSeek-R1-7B-GuardModel Class=TRACEGUARD (Ours), Detector Variant=DeepSeek-R1-7B-Guard2026.03 | 71.2 | 75.4 | |
| DeepSeek-R1-7BModel Class=Zero-Shot Baselines, Detector Variant=DeepSeek-R1-7B [3]2026.03 | 31.4 | 35.6 | |
| Llama-Guard-3-8BModel Class=Llama-Guard [26], Detector Variant=Llama-Guard-3-8B [26]2026.03 | 0 | — |