PII Masking Health Phi 400K: Multilingual Medical PII Annotations

417,900 entries contain 2,802,316 annotations for personal health information across 30 languages and 37 regions. This dataset, part of the larger PII-Masking-3M corpus, is created by Ai4Privacy for training models to detect and mask sensitive medical data. The dataset was last updated on June 3, 2026.

Use Cases

• Train PII detection models based on the 37 annotation labels mentioned in the description
• Benchmark multilingual NER performance based on the 30 languages covered
• Develop text anonymization tools for global health records based on the dataset's regional coverage
• Fine-tune models for medical text de-identification based on the health and medical information focus

Strengths

• Contains 417,900 text entries
• Includes 2,802,316 individual PII annotations
• Covers 30 languages and 37 regions, suggesting broad applicability

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment

Provenance

Source

Ai4Privacy

Collection Method

Part of the PII-Masking-3M corpus, likely annotated from various sources of personal health information.

Freshness

Last updated 2026-06-03 07:27:00

Geography

Global coverage across 37 regions