ChatGPT gets your prompt before you hit send

Name: ChatGPT gets your prompt before you hit send
Availability: InStock
Author: phico

by phico·Mar 3, 2026·3 points·7 comments

AI Analysis

●●SolidSolve My ProblemDark HorseShip It

Isolated browser extension overlay intercepts keystrokes ChatGPT can't read—but users still trust OpenAI with unmasked context.

Strengths

•Legitimate isolation mechanism: extension context prevents host page scripts from accessing input before masking happens.
•Client-side PII detection (names, emails, IBANs) with token replacement—no server-side processing or data transmission.
•Open source on GitHub with verification instructions, addressing the core trust requirement for privacy tools.

Weaknesses

•False sense of security: masking '[NAME_1]' still sends the full conversation to OpenAI—you're trusting their servers anyway.
•Crowded category: 1Password, Dashlane, and browser password managers already prevent keystroke leakage; Anthropic's Claude has native data handling controls.