OpenAI Releases Privacy Filter Model to Detect and Redact PII
Summary
OpenAI released Privacy Filter, an open-weight model for locally detecting and redacting PII. OpenAI reports F1 scores of 96% and 97.43% on benchmark versions, but cautions that high-sensitivity use requires human review.
Key points
- Privacy Filter is an open-weight model released under the Apache 2.0 license and available on Hugging Face and GitHub.
- It performs token classification in a single pass, can run locally, and supports contexts up to 128,000 tokens.
- The model identifies eight PII categories, including names, contact details, account numbers, and secrets such as passwords and API keys.
- OpenAI reported a 96% F1 score on PII-Masking-300k and 97.43% on a revised dataset.
- OpenAI warns the model may miss uncommon or ambiguous identifiers or over-redact, and recommends human review and domain-specific evaluation for high-sensitivity workflows.
Article Details
- Event Type
- OpenAI released Privacy Filter, an open-weight model for detecting and redacting personally identifiable information in text.
- Impact
- Privacy Filter is designed to reduce exposure of personal data in AI workflows by identifying and masking sensitive information, and can run locally so unfiltered data need not be sent to a server. OpenAI reported 96% F1 on PII-Masking-300k and 97.43% on a revised dataset. The model may miss uncommon identifiers or over- or under-redact; OpenAI recommends human review and domain-specific evaluation and fine-tuning for high-sensitivity uses.
Vendors
Products
ChatGPTSince people tend to paste personal data into AI tools such as ChatGPT, OpenAI has released Privacy Filter, an open-weight model designed to detect and redact personally identifiable information (PII) in text.Privacy FilterOpenAI released Privacy Filter, an open-weight model under the Apache 2.0 license that detects and redacts PII in text and is available on Hugging Face and GitHub.