Malicious Chrome extensions targeting AI conversations

Security researchers have identified malicious Google Chrome extensions designed to intercept and steal user conversations from various AI platforms, including ChatGPT, Claude, Gemini, and DeepSeek.

Initial reports highlighted an extension named ‘Prompt Optimizer – SecondBrain’ (version 2.3.1) installed on approximately 100,000 devices. This extension reportedly bypassed privacy claims by using the Fetch API, XMLHttpRequest, and WebSocket to record full user conversations. The collected data was encrypted with AES-GCM using a server-supplied key for later decryption.

Subsequent findings revealed that a similar threat, titled ‘AI Sidebar with DeepSeek, ChatGPT, Claude and more’, resurfaced in the Chrome Web Store despite being previously removed. Researchers identified a ‘clean and then poison’ strategy, where developers released a legitimate version to regain trust before deploying a version containing malicious code. This version encoded conversation text in Base64 to transmit it to external domains without user knowledge.