Introducing OpenAI Privacy Filter | OpenAI
Short summary
OpenAI released Privacy Filter, an open-weight model that detects and redacts eight categories of personally identifiable information in text using context-aware language understanding rather than simple pattern matching. It runs locally on long inputs in a single pass and uses BIOES span tags for precise masking. The model was built by converting a pretrained language model into a token classifier and trained on a mix of public and synthetic data.
- Uses context-aware language understanding instead of regex, so it distinguishes public figures from private individuals and catches hidden credentials like passwords
- Small enough to run entirely locally, processing long documents in a single pass
- Built by converting a pretrained language model into a bidirectional token classifier trained on public plus synthetic data, with model-assisted annotation and human review filling gaps
- Tags eight PII categories using BIOES span tagging for clean, precise redaction boundaries
Full summary
OpenAI just released Privacy Filter, an open-weight model that spots and redacts personally identifiable information (PII) in text. It’s small, runs entirely on your machine, and handles long documents in one fast pass. Instead of simple regex rules for emails or phone numbers, it uses language understanding to decide what counts as private based on context—so it skips public names but flags a private person’s name, or masks a password even when it’s hidden in a sentence.
Privacy Filter labels text spans in eight categories: private_person, private_address, private_email, private_phone, private_url, private_date, account_number and secret. Account_number catches credit cards, bank account numbers and similar codes. Secret covers passwords, API keys and other credentials. It uses BIOES tagging (begin, inside, outside, end, single) to ensure clean boundaries around each span. That helps you get precise redactions without chopping words in half.
Behind the scenes, OpenAI started by defining a detailed privacy taxonomy. They turned a standard pretrained language model into a bidirectional token classifier, swapping out the language modeling head for a token-classification head. Training mixed public text with synthetic examples to cover tricky formats and contexts. When public data labels fell short, they ran model-assisted annotation and human review to fill gaps. The result is a compact, context-aware tool aimed at high-throughput privacy workflows.