Unsafe Content Generation Evaluation on Internal Safety Collapse (ISC)
0Unsafe Generation Rate (Guard)SafeRedirect
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SafeRedirectModel=GPT-5.22026.04 | 0 | 0 | 0 | 0 | 58.7 | |
| SafeRedirectModel=GLM-52026.04 | 1 | 0 | 0 | 0.3 | 49.7 | |
| SafeRedirectModel=Kimi K2.52026.04 | 3 | 1 | 5 | 3 | 79.3 | |
| SafeRedirectModel=Grok 4.1 Fast2026.04 | 6 | 0 | 6 | 4 | 86.7 | |
| SafeRedirectModel=Average2026.04 | 10 | 3.9 | 10 | 8 | 63.2 | |
| SafeRedirectModel=Claude Sonnet 4.52026.04 | 14 | 0 | 0 | 4.7 | 78 | |
| SafeRedirectModel=Gemini 2.5 Pro2026.04 | 17 | 1 | 33 | 17 | 48 | |
| SPDModel=GLM-52026.04 | 20 | 22 | 20 | 20.7 | — | |
| SafeRedirectModel=MiniMax M2.72026.04 | 29 | 25 | 26 | 26.7 | 42.7 | |
| SPDModel=GPT-5.22026.04 | 35 | 3 | 39 | 25.7 | — | |
| No DefenseModel=GLM-52026.04 | 58 | 29 | 63 | 50 | — | |
| No DefenseModel=Gemini 2.5 Pro2026.04 | 63 | 64 | 68 | 65 | — | |
| SPDModel=Average2026.04 | 63.3 | 51.3 | 50.3 | 55 | — | |
| SPDModel=Gemini 2.5 Pro2026.04 | 66 | 55 | 52 | 57.7 | — | |
| SPDModel=MiniMax M2.72026.04 | 66 | 61 | 51 | 59.3 | — | |
| No DefenseModel=MiniMax M2.72026.04 | 69 | 71 | 68 | 69.3 | — | |
| No DefenseModel=Average2026.04 | 77 | 61.1 | 75.6 | 71.2 | — | |
| No DefenseModel=GPT-5.22026.04 | 80 | 20 | 76 | 58.7 | — | |
| SPDModel=Claude Sonnet 4.52026.04 | 80 | 59 | 23 | 54 | — | |
| SPDModel=Kimi K2.52026.04 | 85 | 66 | 77 | 76 | — | |
| No DefenseModel=Claude Sonnet 4.52026.04 | 86 | 77 | 85 | 82.7 | — | |
| No DefenseModel=Kimi K2.52026.04 | 91 | 73 | 83 | 82.3 | — | |
| SPDModel=Grok 4.1 Fast2026.04 | 91 | 93 | 90 | 91.3 | — | |
| No DefenseModel=Grok 4.1 Fast2026.04 | 92 | 94 | 86 | 90.7 | — |