PII Arena
HF detector

nym-base+homoglyph PII detection benchmark

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

71.53%fully hidden annotations
87.22%detected / overlapped
4.91%extra masking
CPU characters / second

Strengths

Highest complete-masking rates by normalized data type.

Bank accounts & cards81.45%303 / 372
Phone numbers & email81.39%634 / 779
People's names76.41%1,137 / 1,488

Watch list

Lowest observed complete-masking rates; inspect before deployment.

Addresses & locations57.48%815 / 1,418
Passwords, keys & tokens67.36%258 / 383
Network identifiers71.75%259 / 361

Language coverage

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

LanguageDatasetsFully hidden ↑Detected ↑Extra masking ↓
English163.19%96.18%6.30%
Russian271.87%86.86%1.75%

Configuration, source and version

Family
HF
Revision
0ed0242d67a30ca6a83e5f118506ff1e1daf83ca
Eligible datasets
3 / 41
Character F1
0.592
Untouched diagnostic
12.78% · 957 / 7,486
Benchmark version
1.0.3 · 2026-09-09
CPU seconds / 10k chars
Not measured
GPU seconds / 10k chars
0.169

source overlap: nym-ru, nym-en, nym-multi

Results by dataset

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

DatasetGold annotationsFully hidden ↑Detected ↑Extra masking ↓
corrupt-secrets-issues28863.19%96.18%6.30%
corrupt-hivetrace1,66771.33%97.36%0.40%
corrupt-redmadrobot5,53172.03%83.69%1.95%