AI Briefing
KOSign in

Protegrity Releases Open-Weights 278M PII Detection Model and PII-SCOPE Benchmark

·2026.11.04 04:29

Key point

The model is a RoBERTa fine-tune trained on 1.2B synthetic samples and identifies PII based on surrounding context.

Details

  • Protegrity released an open-weights 278M parameter PII detection model and the PII-SCOPE benchmark.
  • The model is a fine-tune of RoBERTa trained on 1.2B samples of synthetically generated text.
  • PII-SCOPE (Synthetic Context and Ontology for PII Extraction) is designed to address issues in public LLM-generated datasets.
  • The model identifies PII instances by index and type based on context, or returns input with PII redacted.
  • Resources are available on Hugging Face and arXiv.
open-sourcebenchmarkpii-detectionrobertahugging-faceprotegrity
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

PyTorchKRMistral Large 4 Achieves Coding Performance Similar to GLM-5.3 with 1T MoE ModelPyTorchKRAnySpeechRecognizer: Swift package supporting open models compatible with Apple Speech APIRedditDatalab releases Surya OCR 2.5, a 689M parameter model with 85.9 OmniParseBench scoreRedditinclusionAI releases ZwZ2 models (27B/9B/4B) based on Qwen3.6-27BHuggingFace BlogHugging Face Releases AnySpeechRecognizer and AnyDecisionModel Swift Packages for Apple PlatformsGeekNewsHerdr releases native desktop app for managing terminals and coding agents