Protegrity-PII
Protegrity's 278M-parameter open-weights encoder that tags 69 PII types in free text, published on Hugging Face.
Overview
Protegrity published Protegrity-PII, an open-weights PII detection model on Hugging Face, on November 6, 2026. It is a 278M-parameter token-level encoder built on XLM-RoBERTa that classifies 69 PII types, including account and card data, device identifiers, health information, credentials and biometric data. On Protegrity's PII-SCOPE benchmark (8,030 records, eight sectors) it reports 0.910 typed recall at 0.902 precision, against 0.465 for AWS Comprehend, 0.387 for OpenAI Privacy Filter and 0.386 for Microsoft Presidio.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- 278M parameters, encoder-only token classifier on XLM-RoBERTa
- 69 PII entity types in free-form text
- 0.910 typed recall at 0.902 precision on PII-SCOPE (company-reported)
- 1M characters in 0.49 seconds on one RTX 6000 Blackwell GPU, about $0.0005 (company estimate)
- Open weights on Hugging Face; license not stated in launch coverage
Use cases
- Redacting PII from text before it is sent to an LLM
- Scanning logs and documents for regulated data
- Self-hosted PII detection where a cloud API is not allowed
Pricing
Free to download as open weights; license terms not stated in launch coverage, so check the Hugging Face model card.
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- 6 Nov 2026
- Record updated
- 7 Nov 2026
- Record last checked
- 7 Nov 2026