Cross-modal attack detection on SkillMutator GPT-5.4 adversarial-oracle (evaluable)
86.8Detection RateGPT-5.4
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5.4Scanner Type=Proprietary LLM2026.06 | 86.8 | |
| GPT-5.4-miniScanner Type=Proprietary LLM2026.06 | 79 | |
| GPT-4o-miniScanner Type=Proprietary LLM2026.06 | 23.7 | |
| PIGuardScanner Type=Prompt-Injection Detector2026.06 | 17.1 | |
| DataSentinelScanner Type=Prompt-Injection Detector2026.06 | 10.5 | |
| Snyk Agent ScanScanner Type=Rule-based / Commercial2026.06 | 9.2 | |
| skill-security-scanScanner Type=Rule-based / Commercial2026.06 | 7.9 | |
| LLM-GuardScanner Type=Prompt-Injection Detector2026.06 | 4 | |
| SkillScanScanner Type=Rule-based / Commercial2026.06 | 1.3 |