Cognitive Poisoning Detection on Cognitive Poisoning Detection 5-fold evaluation (test)
4.2AMRVISTA-Guard
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VISTA-GuardBackbone=Mistral-7B2026.05 | 4.2 | 95.8 | 5 | 95 | 84.2 | |
| VISTA-GuardBackbone=Llama-3.1-8B2026.05 | 4.6 | 95.4 | 18.8 | 81.2 | 62.6 | |
| VISTA-GuardBackbone=OLMo-2-7B2026.05 | 6.6 | 93.4 | 15.1 | 84.9 | 64.2 | |
| VISTA-GuardBackbone=Qwen3-8B2026.05 | 7.6 | 92.4 | 11.6 | 88.4 | 67.6 | |
| VISTA-GuardBackbone=Qwen2.5-7B2026.05 | 9.6 | 90.4 | 10.6 | 89.4 | 64.9 | |
| GPT-5.4Mode=Full (trajectory + params), Context=Zero-shot judge2026.05 | 22.2 | 77.8 | 80.4 | 19.6 | 0 | |
| GPT-5.4Mode=Features only (10 scalars), Context=Zero-shot judge2026.05 | 25.2 | 74.8 | 80.9 | 19.1 | 0 |