PII Arena
HF detector

ner-ru-gherman+homoglyph PII detection benchmark

Measured on 3 of 41 retained datasets under the same frozen scoring and masking protocol.

23.39%fully hidden annotations
36.13%detected / overlapped
0.17%extra masking
CPU characters / second

Strengths

Highest complete-masking rates by normalized data type.

People's names73.25%1,090 / 1,488
Addresses & locations46.12%654 / 1,418
Bank accounts & cards0.27%1 / 372

Watch list

Lowest observed complete-masking rates; inspect before deployment.

Network identifiers0.00%0 / 361
Passwords, keys & tokens0.00%0 / 383
Phone numbers & email0.13%1 / 779

Language coverage

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

LanguageDatasetsFully hidden ↑Detected ↑Extra masking ↓
English10.00%0.00%0.12%
Russian224.33%37.58%0.29%

Configuration, source and version

Family
HF
Revision
fc6b2c5a2c5d7c82da7416f7fd9c055159bbb984
Eligible datasets
3 / 41
Character F1
0.318
Untouched diagnostic
63.87% · 4,781 / 7,486
Benchmark version
1.0.3 · 2026-09-09
CPU seconds / 10k chars
Not measured
GPU seconds / 10k chars
0.172

Results by dataset

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

DatasetGold annotationsFully hidden ↑Detected ↑Extra masking ↓
corrupt-secrets-issues2880.00%0.00%0.12%
corrupt-hivetrace1,6671.38%28.43%0.37%
corrupt-redmadrobot5,53131.24%40.34%0.28%