OpenAI ·

OpenAI releases Privacy Filter, an open-weight model for masking personally identifiable information in text, with 1.5B total and 50M active parameters

Our state of the art model for masking personally identifiable information (PII) in text

OpenAI releases Privacy Filter, an open-weight model for masking personally identifiable information in text, with 1.5B total and 50M active parameters

Lead Source

How this story grew

Coverage · 0 Discussion · 0
Apr 22Apr 24

More

Help Net Security: Help Net Security
GitHub: GitHub
VentureBeat: VentureBeat
Decrypt: Decrypt
Bloomberg Law: Bloomberg Law

Discussion

TechSnif Coverage

OpenAI Drops Open-Weight Model to Scrub Your Personal Data

OpenAI's new Privacy Filter model masks personally identifiable information in text using just 50M active parameters.

OpenAI just open-sourced a purpose-built model for stripping personally identifiable information from text. The Privacy Filter is an open-weight release, meaning developers can inspect and deploy it freely.

The model packs 1.5 billion total parameters but runs on only 50 million active ones. That sparse architecture means it's lightweight enough for practical deployment while still delivering what OpenAI calls state-of-the-art PII masking performance.

It's a notable move from a company not exactly known for giving things away. Open-weight releases from OpenAI remain rare, making this a meaningful signal about how the company views privacy tooling — as infrastructure worth sharing rather than monetizing directly.

For developers handling sensitive text data, this could become a go-to preprocessing step. Small footprint, specific job, open weights. That's a useful combination.