SQL Agent data leakage evaluation on Employee Toy
100BA (Benign Accuracy)gpt-4.1-mini
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| gpt-4.1-miniSetup=Orchestrator with SQL Agent, Attack Type=Explicit Attacks2026.02 | 100 | 84 | 6 | |
| gpt-4.1Setup=Orchestrator with SQL Agent, Attack Type=Explicit Attacks2026.02 | 100 | 75.8 | 23 | |
| o4-miniSetup=Orchestrator with SQL Agent, Attack Type=Explicit Attacks2026.02 | 100 | 90.6 | 500 | |
| claude-sonnet-4Setup=Orchestrator with SQL Agent, Attack Type=Explicit Attacks2026.02 | 100 | 93.6 | — | |
| gemini-2.5-flashSetup=Orchestrator with SQL Agent, Attack Type=Explicit Attacks2026.02 | 100 | 75.4 | 17 | |
| gpt-4.1-mini2026.02 | 100 | 87.2 | 7 | |
| gpt-4.12026.02 | 100 | 77.8 | 42 | |
| o4-mini2026.02 | 100 | 90.4 | — | |
| claude-sonnet-42026.02 | 100 | 95.4 | — | |
| gemini-2.5-flash2026.02 | 100 | 76.4 | 18 |