List your company

SQL Generation on EHR SQL

90.7AccuracySkillTrojan
65.32471.91278.585.088Apr 8, 2026

Evaluation results

MethodLinks
SkillTrojanModel=Claude-Sonnet4.52026.0490.764.2
SkillTrojanModel=GPT-5.2-1211-Global2026.0489.397.2
SkillTrojanModel=Qwen3-Max2026.0486.674.7
Non-attackModel=Claude-Sonnet4.52026.0486.50
SkillTrojanModel=Claude-Haiku-4.52026.0482.757.3
Non-attackModel=Qwen3-Max2026.0482.70
AgentPoisonModel=Claude-Haiku-4.52026.0475.554.8
AgentPoisonModel=GPT-4o-mini-0718-Global2026.0474.853.7
AgentPoisonModel=Claude-Sonnet4.52026.0474.156.2
AgentPoisonModel=Qwen3-Max2026.0473.457.6
Non-attackModel=Claude-Haiku-4.52026.0473.10
Non-attackModel=GPT-5.2-1211-Global2026.04730
AgentPoisonModel=GPT-5.2-1211-Global2026.0472.958.3
BadChainModel=Claude-Haiku-4.52026.0472.628.9
BadChainModel=GPT-4o-mini-0718-Global2026.0471.933.7
Non-attackModel=GPT-4o-mini-0718-Global2026.0471.60
BadChainModel=Claude-Sonnet4.52026.0471.522.7
BadChainModel=Qwen3-Max2026.0470.914.2
GCGModel=GPT-4o-mini-0718-Global2026.0470.830.2
BadChainModel=GPT-5.2-1211-Global2026.0470.88.3
GCGModel=Claude-Haiku-4.52026.0470.132.9
GCGModel=GPT-5.2-1211-Global2026.0470.145.8
GCGModel=Qwen3-Max2026.0469.842.1
AutoDANModel=Claude-Sonnet4.52026.0469.734.9
GCGModel=Claude-Sonnet4.52026.0469.637.8
AutoDANModel=Claude-Haiku-4.52026.0468.639.4
SkillTrojanModel=GPT-4o-mini-0718-Global2026.0468.554.3
AutoDANModel=GPT-5.2-1211-Global2026.0468.427.4
AutoDANModel=Qwen3-Max2026.0467.930.8
CPAModel=GPT-5.2-1211-Global2026.0467.951.1
CPAModel=Claude-Sonnet4.52026.0467.844.6
AutoDANModel=GPT-4o-mini-0718-Global2026.0467.442.1
CPAModel=Qwen3-Max2026.0467.248.7
CPAModel=Claude-Haiku-4.52026.0466.941
CPAModel=GPT-4o-mini-0718-Global2026.0466.338.5