PII Arena
PPLX detector

pplx+cpu-int8 PII detection benchmark

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

93.12%fully hidden annotations
96.34%detected / overlapped
13.53%extra masking
CPU characters / second

Strengths

Highest complete-masking rates by normalized data type.

Dates & times99.95%1,860 / 1,861
Logins & usernames99.82%8,775 / 8,791
Other sensitive attributes99.55%223 / 224

Watch list

Lowest observed complete-masking rates; inspect before deployment.

Organizations27.61%1,261 / 4,568
Addresses & locations93.27%8,308 / 8,907
Customer & employee IDs93.62%1,703 / 1,819

Language coverage

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

LanguageDatasetsFully hidden ↑Detected ↑Extra masking ↓
English481.98%86.30%16.89%
Russian495.83%99.17%4.23%
Multilingual198.80%99.00%

Configuration, source and version

Family
PPLX
Revision
f1f90a53823f5df0a1344c1e137d9fffdaab54d6
Eligible datasets
9 / 41
Character F1
0.698
Untouched diagnostic
3.66% · 2,639 / 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-echr3,83047.91%57.31%
secrets-issues28881.25%100.00%18.81%
redmadrobot5,51687.71%94.31%4.07%
nemotron-pii9,39192.91%95.55%13.67%
ameau012,24294.47%95.32%6.98%
nym-ru9,49795.61%99.51%
synth-ru-tickets32,09997.19%99.91%
hivetrace1,66797.90%99.16%5.27%
kiji-multi7,56598.80%99.00%