AnonimiNER
Specialized AI model for automatic identification and anonymization of personal data in any Polish-language texts. Understands context, distinguishes personal data from medical terms, and operates 100% on-premise.
Model Description
AnonimiNER is a specialized AI model based on NER (Named Entity Recognition) technology, designed for automatic identification and anonymization of personal data in any Polish-language texts. The model detects names, surnames, PESEL numbers, addresses, phone numbers, and other sensitive data — regardless of whether you are processing medical documentation, court records, HR files, or customer conversation transcripts.
Unlike traditional tools based on rigid rules and regular expressions, AnonimiNER understands sentence context and decides whether a given word is personal data. The model knows that "patient Kowalski" is data to anonymize, but "Parkinson's disease" or "Asperger's syndrome" are medical terms that should remain intact. It can also distinguish a PESEL number from an invoice number or payment reference, even though all are digit strings — because it analyzes meaning, not just patterns.
AnonimiNER is a Small Language Model — intelligent enough to understand language nuances, yet lightweight enough to run fast on standard server hardware without costly GPU infrastructure. Document processing takes milliseconds, enabling batch anonymization of thousands of files.
With on-premise deployment, data never leaves your infrastructure, ensuring full GDPR compliance and eliminating the risk of information leaks.
94.92% Recall — nearly every piece of personal data detected
In the strict recall benchmark, AnonimiNER scored 94.92% (vs 49.44% for Presidio). This means the model misses only ~5% of personal data, while Presidio loses more than half. For PESEL and phone numbers, recall is 100%.
Recognized Personal Data
Anonymization Methods
AnonimiNER vs Presidio
Independent benchmark on 200 clinical cases · February 2026
Recall - how much personal data was detected
Each undetected element is a potential PII leak
Accuracy for individual data types
| Data type | AnonimiNER | |
|---|---|---|
| Recall | F1 | |
| Names and surnames | 95.34% | 83.96% |
| PESEL numbers | 100.00% | 100.00% |
| Phone numbers | 100.00% | 99.22% |
| Home addresses * | 87.70% | 84.25% |
| Email addresses | 97.96% | 98.97% |
* Presidio detects address fragments (high recall), but does not match full spans (low F1).
Why this matters
- •Strict Recall 94.92% — AnonimiNER misses only ~5% of personal data. Presidio loses more than half (recall 49.44%).
- •PESEL: 100% recall vs 25.81% for Presidio. Phone numbers: 100% vs 20.31%. Presidio misses 3 out of 4 PESEL numbers and 4 out of 5 phone numbers.
- •High recall with high F1 (90.20%) means the model not only finds data but also precisely matches its boundaries.
Applications in DLP (Data Loss Prevention) Systems
AnonimiNER is a key component of Data Loss Prevention (DLP) strategy in organizations processing personal data. Unlike traditional DLP systems that only block the transmission of sensitive information, AnonimiNER enables safe data use through automatic anonymization.
How AnonimiNER supports your DLP strategy:
AI Protection Layer
AnonimiNER acts as a protection layer between your data and external AI systems:
Pacjent Jan Kowalski, PESEL 80010112345, zamieszkały ul. Zielona 12, Warszawa, tel. 123-456-789 zgłosił się na wizytę kontrolną z powodu nadciśnienia tętniczego.
Pacjent [OSOBA], PESEL [PESEL], zamieszkały [ADRES], tel. [TELEFON] zgłosił się na wizytę kontrolną z powodu nadciśnienia tętniczego.
Typical use cases:
- •Summarizing documents via LLM without revealing personal data
- •Translating documents containing sensitive data
- •Generating reports based on anonymized sources
- •Sentiment analysis of customer/employee feedback
- •Extracting information from documents (without retaining PII in external systems)
Business Benefits
FAQ – Frequently Asked Questions
API Integration
Models are available as a REST API service or for local deployment within hospital infrastructure. We provide documentation and integration support.
curl -X POST https://api.medalion.tech/v1/anonymize \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"text": "Patient Jan Kowalski, PESEL: 80010112345, tel: 123-456-789",
"method": "mask",
"entities": ["PERSON", "PESEL", "PHONE"]
}'Anonymization, DLP, compliance audit — one tool
AnonimiNER works for medical documentation anonymization, data leak prevention (DLP), GDPR audits, and safe dataset sharing for analytics and AI training. On-premise deployment or API.