The dataset is currently empty. Upload or create new data files. Then, you will be able to explore them in the Dataset Viewer.
Double Logging to Detect an Unaligned LLM
Can comparing a model's own action log with an automatic system log detect harmful behaviour? The detector sees only the two logs and is scored on detection rate and false alarms.
Status: planned, not yet released. The benchmark scenarios are still in review and no study runs have been made. This repository holds no data yet; it is a placeholder for the future release.
| Data | None yet |
| Code | GitHub |
| Earlier study | INSIDER_LLM_DETECTION_BENCHMARK (archived) |
- Downloads last month
- 20