Kiji Proxy: Detect/mask private infos in requests to AI services (github.com)

575 Lab, part of Dataiku's Open Source Office, has launched Kiji Proxy, a tool designed to enhance privacy for users of AI services by automatically detecting and masking personally identifiable information (PII) in API requests. This smart privacy layer recognizes 26 types of PII—like emails and credit card numbers—through a machine learning model, replacing sensitive data with realistic dummy values before the data is sent to external servers, such as those operated by OpenAI or Anthropic. Kiji ensures that original information is seamlessly restored in responses, allowing applications to function normally without any code changes. The significance of Kiji Proxy lies in its robust approach to data privacy, which addresses growing concerns over data leaks and regulatory compliance (GDPR, HIPAA, CCPA) in AI applications. By processing data locally without external API dependencies, Kiji offers fast, sub-100ms real-time PII detection and protection directly from a transparent proxy setup, available on both macOS and Linux. Its browser integration and Chrome extension make it user-friendly, encouraging safer use of AI tools in various contexts, including customer support and data analysis, while safeguarding sensitive information against inadvertent exposure.