OpenAI Releases Privacy Filter: An Open-Weight Model for Detecting and Redacting PII at State-of-the-Art Accuracy

OpenAI released the Privacy Filter on April 22, 2026 — an open-weight model specifically designed to detect and redact personally identifiable information (PII) from text with accuracy that exceeds previous open-source alternatives on standard PII detection benchmarks. The model ships open-weight, meaning organizations can run it on their own infrastructure without sending sensitive text to an external API — a critical requirement for healthcare, legal, financial, and HR workflows that handle regulated personal data. For enterprise automation pipelines that process customer communications, intake forms, support tickets, or employee records, Privacy Filter provides a production-ready PII scrubbing layer that can run inline before data reaches downstream AI models, reducing the risk of accidentally exposing personal data through model inputs or outputs. The open-weight release puts professional-grade PII detection within reach of any engineering team, including those operating in air-gapped or compliance-restricted environments.