SQL Generation on EHR SQL
90.7AccuracySkillTrojan
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SkillTrojanModel=Claude-Sonnet4.52026.04 | 90.7 | 64.2 | |
| SkillTrojanModel=GPT-5.2-1211-Global2026.04 | 89.3 | 97.2 | |
| SkillTrojanModel=Qwen3-Max2026.04 | 86.6 | 74.7 | |
| Non-attackModel=Claude-Sonnet4.52026.04 | 86.5 | 0 | |
| SkillTrojanModel=Claude-Haiku-4.52026.04 | 82.7 | 57.3 | |
| Non-attackModel=Qwen3-Max2026.04 | 82.7 | 0 | |
| AgentPoisonModel=Claude-Haiku-4.52026.04 | 75.5 | 54.8 | |
| AgentPoisonModel=GPT-4o-mini-0718-Global2026.04 | 74.8 | 53.7 | |
| AgentPoisonModel=Claude-Sonnet4.52026.04 | 74.1 | 56.2 | |
| AgentPoisonModel=Qwen3-Max2026.04 | 73.4 | 57.6 | |
| Non-attackModel=Claude-Haiku-4.52026.04 | 73.1 | 0 | |
| Non-attackModel=GPT-5.2-1211-Global2026.04 | 73 | 0 | |
| AgentPoisonModel=GPT-5.2-1211-Global2026.04 | 72.9 | 58.3 | |
| BadChainModel=Claude-Haiku-4.52026.04 | 72.6 | 28.9 | |
| BadChainModel=GPT-4o-mini-0718-Global2026.04 | 71.9 | 33.7 | |
| Non-attackModel=GPT-4o-mini-0718-Global2026.04 | 71.6 | 0 | |
| BadChainModel=Claude-Sonnet4.52026.04 | 71.5 | 22.7 | |
| BadChainModel=Qwen3-Max2026.04 | 70.9 | 14.2 | |
| GCGModel=GPT-4o-mini-0718-Global2026.04 | 70.8 | 30.2 | |
| BadChainModel=GPT-5.2-1211-Global2026.04 | 70.8 | 8.3 | |
| GCGModel=Claude-Haiku-4.52026.04 | 70.1 | 32.9 | |
| GCGModel=GPT-5.2-1211-Global2026.04 | 70.1 | 45.8 | |
| GCGModel=Qwen3-Max2026.04 | 69.8 | 42.1 | |
| AutoDANModel=Claude-Sonnet4.52026.04 | 69.7 | 34.9 | |
| GCGModel=Claude-Sonnet4.52026.04 | 69.6 | 37.8 | |
| AutoDANModel=Claude-Haiku-4.52026.04 | 68.6 | 39.4 | |
| SkillTrojanModel=GPT-4o-mini-0718-Global2026.04 | 68.5 | 54.3 | |
| AutoDANModel=GPT-5.2-1211-Global2026.04 | 68.4 | 27.4 | |
| AutoDANModel=Qwen3-Max2026.04 | 67.9 | 30.8 | |
| CPAModel=GPT-5.2-1211-Global2026.04 | 67.9 | 51.1 | |
| CPAModel=Claude-Sonnet4.52026.04 | 67.8 | 44.6 | |
| AutoDANModel=GPT-4o-mini-0718-Global2026.04 | 67.4 | 42.1 | |
| CPAModel=Qwen3-Max2026.04 | 67.2 | 48.7 | |
| CPAModel=Claude-Haiku-4.52026.04 | 66.9 | 41 | |
| CPAModel=GPT-4o-mini-0718-Global2026.04 | 66.3 | 38.5 |