Show HN: Reverse engineering user queries for ChatGPT/Gemini
A new open-source tool, AI Visibility Audit, reverse-engineers user queries sent to AI platforms like ChatGPT and Gemini by analyzing network traffic. It aims to reveal how these services handle and potentially expose user data through their APIs.
Background
- This GitHub project is a tool that intercepts and logs what users type into AI assistants like ChatGPT and Gemini before the query is sent to the server. It works by hooking into the browser's JavaScript runtime to capture the text as it's being typed or pasted into the chat input field.
- The "Reverse engineering" in the title refers to figuring out exactly how these AI chat interfaces handle user input internally (e.g., how they format, store, and transmit the text), not to any form of breaking encryption or accessing other users' data.
- This is relevant to developers, security researchers, and privacy advocates. The project raises awareness that AI chat apps, even when running in a browser, are not black boxes — client-side code can be inspected and modified, and user input is visible locally before it's sent anywhere.
- It also has implications for privacy auditing: such tools can verify whether an AI service is sending keystroke data, metadata, or drafts before a user explicitly presses "send," which some privacy-conscious users worry about.