Redact: New Chrome Extension Blocks Credential and PII Leaks to AI Chat Services
Key Takeaways
- ▸Detects credentials and PII before they reach AI chat services using on-device neural network inference (~150ms per detection)
- ▸Privacy-first design: all processing happens locally in the browser with zero network calls, telemetry, or external dependencies
- ▸Users get real-time control: choose to block, warn, or send for each sensitive data type with customizable defaults tuned for engineers
Summary
Redact, a new open-source Chrome extension, helps prevent accidental leaks of credentials, passwords, and personally identifiable information (PII) to large language model (LLM) chat services like ChatGPT, Claude, Gemini, Perplexity, Grok, and others. The tool works by intercepting pastes on supported chat sites and using on-device neural network inference to detect sensitive data before it reaches the AI service. Users receive an option to block, warn, or proceed with sending the data, giving them full control over what information reaches these services.
The extension prioritizes privacy and security through a privacy-first architecture: all detection runs locally in the browser using a Web Worker, with no network calls or telemetry. The model file is bundled directly in the extension package, and the entire source code is auditable on GitHub under a PolyForm Noncommercial license. Detection happens in approximately 150 milliseconds, making it practical for real-time use. The tool is free in the Chrome Web Store and currently supports all major LLM chat platforms.
This addresses a critical vulnerability in how users interact with AI services: chat transcripts are stored on provider servers indefinitely, making the paste box a significant credential leak risk. By running detection entirely on-device with no external calls, Redact provides users with a practical defense against accidentally compromising sensitive information like API keys, database credentials, passwords, and personal data.
- Free and open source on Chrome Web Store with auditable GitHub source code; runs on all major LLM chats
- Addresses the fundamental risk that chat transcripts stored on AI provider servers cannot be recovered or deleted once leaked
Editorial Opinion
The proliferation of AI chat services has created a new security blind spot: users reflexively paste sensitive data into chat boxes without realizing those transcripts live permanently on third-party servers. Redact addresses this gap elegantly by shifting detection to the client side and running entirely in-browser, putting privacy back in users' hands without requiring trust in new infrastructure. In an era where AI adoption outpaces security awareness, privacy tools like this are essential bridges between user behavior and the realities of data persistence.
