National Design Studio开源发布了Rampart，一款在浏览器本地运行的个人信息识别与脱敏工具1。该系统旨在防止用户在与聊天机器人交互时无意中泄露姓名、社保号、地址等敏感个人信息到远程服务器1。
Rampart采用双层处理架构，结合确定性正则表达式规则与MiniLM小语言模型来识别和脱敏个人信息1。系统能够检测社保号、信用卡号、电话号码、银行账户、电子邮件、IP地址和政府身份证号等多种信息类型，并在消息发送前进行实时处理1。所有数据处理均在浏览器端完成，无需涉及远程服务器1。相比之下，OpenAI Privacy Filter体积约2.8GB且下载需时约38分钟，而Rampart作为轻量级方案显著降低了系统开销1。
该工具支持英文、西班牙文、法文、德文、意大利文、葡萄牙文和荷兰文共七种语言1，已在OpenPII 1.5M数据集的3万行测试集上进行基准测试1。Rampart目前处于Alpha阶段，定位为管理AI聊天体验中个人信息保护的第一道防线1。
National Design Studio has open-sourced Rampart, a privacy protection system designed to prevent users from accidentally leaking personal information when interacting with chatbots 1. The tool operates entirely within the browser, automatically redacting sensitive data such as Social Security numbers, credit card details, phone numbers, bank account information, email addresses, IP addresses, and government IDs before messages are sent to remote servers 1.
Rampart employs a dual-layer approach combining deterministic regular expression rules with the MiniLM small language model to identify and redact personal information in real time 1. The system currently supports seven languages including English, Spanish, French, German, Italian, Portuguese, and Dutch 1. By processing all data locally on the user's device rather than relying on external servers, Rampart offers a lightweight alternative to existing solutions—notably OpenAI's Privacy Filter, which requires approximately 2.8 gigabytes of storage and roughly 38 minutes to download 1. The tool is currently in Alpha stage and functions as a first line of defense for managing personal information exposure in AI chat interactions, having been evaluated against over 30,000 test cases from the OpenPII 1.5M dataset across Latin-script languages 1.
评论
还没有评论，欢迎留下第一条。