OpenAI releases Privacy Filter, a free tool that strips personal data from business files before they touch the cloud

OpenAI launched Privacy Filter today on Hugging Face (a code-sharing platform) as open-source software. The tool is a 1.5-billion-parameter model—a smaller AI trained to spot and remove names, emails, phone numbers, and other sensitive information from documents and datasets. Anyone can download and run it for free on their laptop or inside a web browser.

Unlike sending files to the cloud first and hoping a company protects them, Privacy Filter works on your device before data ever leaves—a 'privacy-by-design' approach. It understands context (so it catches medical records, financial info, and company secrets, not just raw patterns), and it's small enough to run offline without needing expensive servers, making it practical for enterprises handling thousands of documents daily.

For compliance teams, legal departments, and companies training AI models internally, this eliminates the painful choice between 'risk exposing employee/customer data to third parties' or 'manually review thousands of files.' Banks, healthcare firms, and SaaS companies using ChatGPT or other AI services can now automatically scrub their data before uploading it, reducing breach risk and regulatory violations (GDPR, HIPAA, SOC 2 audits).

The tool is free and available now on Hugging Face under the Apache 2.0 license—no signup, no quota, no vendor lock-in. This marks OpenAI's return to open-source after years of focusing on closed, paid models like ChatGPT.