PII Arena
HF detector

ru-legal-ner+sent300 PII detection benchmark

Measured on 11 of 41 retained datasets under the same frozen scoring and masking protocol.

43.64%fully hidden annotations
66.95%detected / overlapped
10.02%extra masking
CPU characters / second

Strengths

Highest complete-masking rates by normalized data type.

People's names76.91%18,728 / 24,350
Phone numbers & email72.19%4,149 / 5,747
Dates & times62.06%373 / 601

Watch list

Lowest observed complete-masking rates; inspect before deployment.

Network identifiers3.66%640 / 17,498
Other sensitive attributes15.75%140 / 889
Customer & employee IDs29.01%686 / 2,365

Language coverage

The overall leader can differ from the best choice for one language slice.

LanguageDatasetsFully hidden ↑Detected ↑Extra masking ↓
English627.56%46.95%10.78%
Russian547.40%71.62%5.00%

Configuration, source and version

Family
HF
Revision
c0313a42ac147ccc38f5c3b0b3e77cab53208683
Eligible datasets
11 / 41
Character F1
0.512
Untouched diagnostic
33.05% · 30,513 / 92,332
Benchmark version
1.0.3 · 2026-09-09
CPU seconds / 10k chars
Not measured
GPU seconds / 10k chars
0.125

Results by dataset

Complete masking is primary; detection shows whether any character overlapped.

DatasetGold annotationsFully hidden ↑Detected ↑Extra masking ↓
tab-echr3,8307.73%26.27%
dialogpii-en3,08716.65%27.31%2.33%
leak-museum10124.75%48.51%7.87%
synth-ru-tickets32,09934.37%78.63%
nemotron-pii9,39135.14%57.18%21.99%
secrets-issues28854.17%82.64%10.45%
nerel24,36955.36%62.53%
factrueval7,96657.18%60.36%
redact-ru9,23061.37%79.87%19.32%
jayguard1,19561.67%80.25%4.64%
creddata77667.65%89.82%12.76%