Introducing OpenAI Privacy Filter

OpenAI
OpenAI has released Privacy Filter, an open-weight, locally-runnable model designed to efficiently detect and redact personally identifiable information in unstructured text.

Summary

OpenAI has introduced Privacy Filter, an open-weight, 1.5B parameter model designed to detect and redact personally identifiable information (PII) from text. The model offers high-throughput, context-aware detection for eight categories, including contact details, account numbers, and secrets like API keys. By running locally, it ensures that sensitive data does not need to be transmitted externally, providing a secure, efficient solution for developers. While it achieves state-of-the-art performance on benchmarks, OpenAI emphasizes that it is a tool for a broader privacy-by-design strategy rather than a total anonymization solution, recommending human oversight in high-stakes environments.

(Source:OpenAI)