eu-pii-safeguard

<p align="center"> <img src="https://tabularis.ai/pii_top_pic.png" alt="ModernFinBERT" width="512"> </p>

<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/Discord%20button.png" width="200"/>

Multilingual Privacy Filter

Multilingual PII Detection Model for European Languages

A state-of-the-art multilingual model for detecting Personally Identifiable Information (PII) across 26 European languages (all EU official languages). It is designed for GDPR compliance, privacy-preserving AI applications, and secure handling of sensitive data in multilingual settings. This model enables enterprises, researchers, and data protection teams to identify and safeguard PII with high accuracy (≈98%) across diverse European contexts.

🎯 Model Performance

• Global F1 Score: 97.02%
• 26 Languages Supported
• 42 PII Entity Types
• Consistent 95%+ F1 across all languages

🌍 Supported Languages

🇧🇬 Bulgarian • 🇨🇿 Czech • 🇩🇰 Danish • 🇩🇪 German • 🇬🇷 Greek • 🇬🇧 English • 🇪🇸 Spanish • 🇪🇪 Estonian • 🇫🇮 Finnish • 🇫🇷 French • 🇮🇪 Irish • 🇭🇷 Croatian • 🇭🇺 Hungarian • 🇮🇹 Italian • 🇱🇹 Lithuanian • 🇱🇻 Latvian • 🇲🇹 Maltese • 🇳🇱 Dutch • 🇵🇱 Polish • 🇵🇹 Portuguese • 🇷🇴 Romanian • 🇷🇺 Russian • 🇸🇰 Slovak • 🇸🇮 Slovenian • 🇸🇪 Swedish • 🇺🇦 Ukrainian

🔍 Detected PII Types

• Personal: First/Last/Middle Names, Age, Gender, Ethnicity
• Contact: Email, Phone, Address, City, Country, Postal Code
• Financial: Credit Card, IBAN, Account Numbers, Salary
• Identity: National ID, Passport, Driver License, Tax ID
• Health: Medical Conditions, Health Insurance ID
• Digital: IP Address, MAC Address, URL, Username, Password
• And more: 42 total entity types

🚀 Quick Start

from transformers import AutoTokenizer, AutoModelForTokenClassification
import torch

# Load model and tokenizer
model_name = "tabularisai/eu-pii-safeguard"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)

# Example text (French)
text = "Bonjour, je suis Marie Dubois, email: [email protected]"

# Tokenize and predict
inputs = tokenizer(text, return_tensors="pt", truncation=True)
with torch.no_grad():
    outputs = model(**inputs)
    predictions = torch.argmax(outputs.logits, dim=-1)

# Get predictions
tokens = tokenizer.convert_ids_to_tokens(inputs["input_ids"][0])
predicted_labels = [model.config.id2label[pred.item()] for pred in predictions[0]]

print("Detected PII:")
for token, label in zip(tokens, predicted_labels):
    if label != "O":
        print(f"  {label}: {token}")

📊 Performance by Language

💼 Use Cases

• 🔒 Data Privacy: Automatically detect and anonymize PII before processing
• ⚖️ GDPR Compliance: Ensure regulatory compliance across EU markets
• 🛡️ Security: Prevent data breaches by identifying sensitive information
• 📊 Data Governance: Audit and catalog personal data in multilingual datasets

🏗️ Model Architecture

• Base Model: XLM-RoBERTa-large
• Task: Token Classification
• Labels: 74 (B-/I- format for 42 entity types)
• Max Length: 256 tokens

🔄 Community Feedback

We're actively seeking feedback from the community! Please:

• 🐛 Report issues or edge cases
• 💡 Suggest improvements
• 🧪 Share your use cases and results
• 📊 Contribute evaluation on new datasets

🏢 About Tabularis AI

Developed by Tabularis AI - Building privacy-preserving AI solutions for enterprise data protection.

---

For questions, collaborations, or licensing inquiries: [email protected]