PII Arena
HF detector

fef2-secret-ru+cpu-int8 PII detection benchmark

Measured on 9 of 41 retained datasets under the same frozen scoring and masking protocol.

27.12%fully hidden annotations
34.61%detected / overlapped
0.50%extra masking
CPU characters / second

Strengths

Highest complete-masking rates by normalized data type.

Logins & usernames84.51%7,429 / 8,791
Passwords, keys & tokens46.91%790 / 1,684
People's names46.10%5,089 / 11,039

Watch list

Lowest observed complete-masking rates; inspect before deployment.

Network identifiers0.05%9 / 18,902
Dates & times0.48%9 / 1,861
Other sensitive attributes2.68%6 / 224

Language coverage

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

LanguageDatasetsFully hidden ↑Detected ↑Extra masking ↓
English49.59%14.74%0.35%
Russian436.44%45.66%0.92%
Multilingual13.58%4.72%

Configuration, source and version

Family
HF
Revision
52b5b0745aac14f73fcf2ac0f91d9b5001a85ae4
Eligible datasets
9 / 41
Character F1
0.443
Untouched diagnostic
65.39% · 47,144 / 72,095
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.

DatasetGold annotationsFully hidden ↑Detected ↑Extra masking ↓
tab-echr3,8302.06%10.31%
kiji-multi7,5653.58%4.72%
ameau012,24210.04%12.62%0.87%
nemotron-pii9,39111.89%15.97%0.00%
secrets-issues28830.90%49.65%0.25%
synth-ru-tickets32,09932.06%41.53%
nym-ru9,49742.86%52.16%
hivetrace1,66746.49%58.61%0.20%
redmadrobot5,51647.81%54.59%1.03%