PII Arena
HF detector

davlan-mbert PII detection benchmark

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

41.98%fully hidden annotations
48.88%detected / overlapped
0.94%extra masking
3,983CPU characters / second

Strengths

Highest complete-masking rates by normalized data type.

People's names88.69%58,303 / 65,736
Organizations66.78%12,871 / 19,274
Addresses & locations59.53%23,577 / 39,602

Watch list

Lowest observed complete-masking rates; inspect before deployment.

Dates & times0.02%1 / 5,502
Network identifiers0.06%16 / 25,631
Passwords, keys & tokens0.12%8 / 6,870

Language coverage

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

LanguageDatasetsFully hidden ↑Detected ↑Extra masking ↓
English1741.99%49.32%0.95%
Russian1938.91%45.37%0.94%
Multilingual549.83%57.55%0.30%

Configuration, source and version

Family
HF
Revision
e756de7f7b8f64fea0c3d7c3872f1322fab747b1
Eligible datasets
41 / 41
Character F1
0.481
Untouched diagnostic
51.12% · 116,270 / 227,466
Benchmark version
1.0.3 · 2026-09-09
CPU seconds / 10k chars
2.511
GPU seconds / 10k chars
0.147

Results by dataset

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

DatasetGold annotationsFully hidden ↑Detected ↑Extra masking ↓
corrupt-secrets-issues2880.00%1.04%0.92%
creddata7760.00%0.13%1.19%
leaky-repo950.00%0.00%0.67%
secrets-issues2880.00%1.04%0.70%
synth-env-configs3920.00%0.00%0.07%
synth-secrets-en5810.00%0.00%0.01%
synth-secrets-ru5810.00%0.00%0.05%
secrets-rules7460.13%0.13%0.80%
arthur-passwords2800.71%0.71%5.55%
leak-museum1010.99%0.99%0.79%
rubai-ru3,4584.05%55.12%
synth-ru-tickets32,0998.24%11.70%
corrupt-hivetrace1,66711.58%22.44%0.57%
synth-wiki-tables13,09413.64%14.40%
hivetrace1,66714.10%24.60%0.18%
synth-jira-comments2,70520.00%20.78%
alexen21,26120.86%44.89%10.05%
multiconer-ru1,20824.17%37.00%2.01%
nym-ru9,49728.46%35.91%
ameau012,24229.17%33.76%3.48%
privy1,89930.28%39.60%0.11%
nym-multi8,16931.02%35.76%0.94%
russian-pii-66k4,80531.26%36.27%
corrupt-redmadrobot5,53132.60%36.18%0.73%
nym-en3,58435.60%41.02%0.53%
redmadrobot5,51638.23%41.15%0.53%
tonicai2,41739.47%44.68%0.05%
gretel-multi4,48442.28%60.66%0.29%
nemotron-pii9,39142.95%47.69%0.00%
redact-ru9,23049.02%58.19%6.54%
alrosait1,86252.63%96.40%3.97%
redact-multi22,86453.74%61.69%
tab-echr3,83053.81%87.10%
dialogpii-multi10,34554.99%64.60%
jayguard1,19555.23%75.40%1.37%
kiji-multi7,56555.72%57.05%
scanpatch8,70855.81%60.51%1.12%
kiji-en7,62356.33%57.96%
dialogpii-en3,08762.97%72.92%2.33%
nerel24,36983.56%88.76%
factrueval7,96693.96%95.43%