OpenAI published Privacy Filter on April 23 under the Apache 2.0 license — a 1.5B-parameter model (50M active) that detects and redacts personally identifiable information entirely on-device so sensitive data never reaches the cloud. The model covers eight PII categories (names, addresses, emails, phone numbers, URLs, dates, account numbers, secrets) and achieves a 96% F1 score on the PII-Masking-300k benchmark with a 128K-token context window. Code and weights are available on GitHub and Hugging Face. OpenAI cautioned the model may under-redact ambiguous references in high-sensitivity domains like legal and medical workflows.