Vulnerability Attack Analysis on WebGenBench Social Media target-domain and cross-domain 1.0
80.65ASR (CodeQL)GPT-5.2
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-5.2Evaluation Protocol=Cross-domain2026.02 | 80.65 | 87.5 | 80.65 | 80 | |
| GPT-5.2Evaluation Protocol=Held-out (target-domain)2026.02 | 80 | 75 | 70 | 52.5 | |
| Claude-4.5 OpusEvaluation Protocol=Cross-domain2026.02 | 80 | 87.5 | 80 | 87.5 | |
| Claude-4.5 OpusEvaluation Protocol=Held-out (target-domain)2026.02 | 75 | 80 | 75 | 77.5 | |
| ComposerEvaluation Protocol=Held-out (target-domain)2026.02 | 75 | 75 | 50 | 75 | |
| GrokEvaluation Protocol=Held-out (target-domain)2026.02 | 75 | 80 | 62.5 | 60.61 | |
| Gemini-3 ProEvaluation Protocol=Cross-domain2026.02 | 75 | 81.25 | 75 | 76.47 | |
| ComposerEvaluation Protocol=Cross-domain2026.02 | 75 | 81.82 | 71.43 | 81.25 | |
| GrokEvaluation Protocol=Cross-domain2026.02 | 75 | 72.22 | 75 | 66.67 | |
| Gemini-3 ProEvaluation Protocol=Held-out (target-domain)2026.02 | 60 | 80 | 60 | 66.67 | |
| Gemini-3 FlashEvaluation Protocol=Cross-domain2026.02 | 57.14 | 70 | 57.14 | 57.14 | |
| Gemini-3 FlashEvaluation Protocol=Held-out (target-domain)2026.02 | 50 | 66.67 | 50 | 66.67 |