OpenAI just dropped Privacy Filter, a small but mighty open-weight model designed to find and redact personally identifiable information (PII) in text. Unlike old-school tools that rely on clunky rules for phone numbers and emails, this one actually understands context — so it can tell the difference between a public figure’s name and a private individual’s info. It runs locally, meaning your sensitive data never has to leave your machine for processing. The model handles long chunks of text in one quick pass, making it ideal for high-volume privacy workflows. OpenAI uses a fine-tuned version internally and claims it beats current benchmarks for PII masking. Developers can run it, fine-tune it, and bake stronger privacy protections into training, logging, and review pipelines. Why it matters: By releasing this as an open-weight model, OpenAI is giving developers a practical tool to build privacy-first AI systems without relying on cloud servers — a big step toward making data protection a default, not an afterthought.