LEAKS detector
deepsecrets PII detection benchmark
Measured on 4 of 41 retained datasets under the same frozen scoring and masking protocol.
15.49%fully hidden annotations
16.88%detected / overlapped
1.15%extra masking
—CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
Passwords, keys & tokens35.38%433 / 1,224
Bank accounts & cards0.00%0 / 169
Documents & identifiers0.00%0 / 609
Watch list
Lowest observed complete-masking rates; inspect before deployment.
Addresses & locations0.00%0 / 176
Phone numbers & email0.00%0 / 390
People's names0.00%0 / 228
Language coverage
The overall leader can differ from the best choice for one language slice.
| Language | Datasets | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| English | 3 | 38.26% | 41.72% | 1.21% |
| Russian | 1 | 0.06% | 0.06% | 0.00% |
Configuration, source and version
- Family
- LEAKS
- Revision
- 2.0.1
- Eligible datasets
- 4 / 41
- Character F1
- 0.325
- Untouched diagnostic
- 83.12% · 2,324 / 2,796
- Benchmark version
- 1.0.3 · 2026-09-09
- CPU seconds / 10k chars
- Not measured
- GPU seconds / 10k chars
- Not measured
Results by dataset
Complete masking is primary; detection shows whether any character overlapped.
| Dataset | Gold annotations | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| hivetrace | 1,667 | 0.06% | 0.06% | 0.00% |
| leaky-repo | 95 | 28.42% | 30.53% | 0.00% |
| secrets-issues | 288 | 34.72% | 44.44% | 0.68% |
| secrets-rules | 746 | 40.88% | 42.09% | 7.19% |