Prompt Exfiltration on Copilot coding agent prompts
0.898Similarity ScoreToolLeak
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ToolLeak2025.09 | 0.898 | 0.512 | |
| Ignore-Remember2025.09 | 0.617 | 0.939 | |
| Remember-the-Start2025.09 | 0.6 | 0.921 | |
| Ignore-Completion-Remember2025.09 | 0.574 | 0.92 | |
| Ignore2025.09 | 0.572 | 0.969 | |
| Completion-Remember2025.09 | 0.566 | 0.921 | |
| Naive2025.09 | 0.563 | 0.967 | |
| Ignore-Completion2025.09 | 0.468 | 0.974 | |
| Completion2025.09 | 0.463 | 0.94 | |
| PLeak2025.09 | 0.164 | 0.978 |