OpenAI Releases Privacy Filter for PII Detection and Redaction

Original: OpenAI Privacy Filter

Why This Matters

Addresses critical privacy needs in AI systems with local processing capabilities

OpenAI released Privacy Filter, an open-weight model for detecting and redacting personally identifiable information in text. The small model offers frontier-level performance, can run locally, and processes long inputs efficiently in a single pass for high-throughput privacy workflows.

OpenAI launched Privacy Filter, a state-of-the-art open-weight model designed to mask personally identifiable information in unstructured text. Unlike traditional PII detection tools that rely on pattern matching, Privacy Filter uses deep language understanding and context awareness to identify subtle personal information. The model achieves state-of-the-art performance on the PII-Masking-300k benchmark and can run locally, ensuring sensitive data never leaves the user's machine. Built as a bidirectional token-classification model with span decoding, it labels input sequences in one pass using a constrained Viterbi procedure. OpenAI uses a fine-tuned version internally and released this tool as part of efforts to support safer AI development by providing practical privacy infrastructure for developers.

Source

openai.com — Read original →