OpenAI Privacy Filter Powers Three Scalable Web Apps

OpenAI has released a Privacy Filter model (1.5B parameters, 50M active, Apache 2.0) that detects PII categories including private_person, private_address, private_email, private_phone, private_url, private_date, account_number, and secret, with a 128,000-token context. Three demo web apps built with Gradio's gr.Server showcase its capabilities: Document Privacy Explorer (ysharma/OPF-Document-PII-Explorer) highlights PII spans in uploaded PDFs or DOCX files with a sidebar filter and summary dashboard; Image Anonymizer (ysharma/OPF-Image-Anonymizer) redacts PII in images with black bars, allowing toggle, drag, and manual annotation on a canvas; SmartRedact Paste (ysharma/OPF-SmartRedact-Paste) provides a pastebin that returns a public redacted URL with placeholders and a private token-gated reveal URL. All apps use @server.api for model endpoints, leveraging Gradio's queue, ZeroGPU allocation, and gradio_client SDK. The model achieves state-of-the-art on the PII-Masking-300k benchmark and is licensed under Apache 2.0.