88Z AI, 23 April — OpenAI has open-sourced a privacy filter model that supports local execution to detect and edit personal identifiable information in text.
At a glance
- The model has 1.5 billion parameters, supports 128K context, and runs locally
- Based on bidirectional token classification, it achieves a 96% F1 score in PII recognition
The update
OpenAI has officially open-sourced the OpenAI Privacy Filter model, designed to help users detect and edit personal identifiable information within text. The model features 1.5 billion parameters, supports a context length of up to 128K, and allows for deployment in local environments. Its core utilizes a bidirectional token classification architecture, capable of accurately identifying eight common types of private data, including names, addresses, emails, phone numbers, accounts, and passwords. On the PII-Masking-300k benchmark, the model achieved an F1 score of 96%, demonstrating excellent capability in recognizing and handling privacy information, thereby providing a new open-source solution for data security.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- FilterSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.