Browser extensions with 8 million users collect extended AI conversations

A troubling discovery has emerged regarding browser extensions boasting over 8 million downloads. These extensions are reportedly capturing extensive conversations from users interacting with AI platforms and are allegedly selling this information for marketing purposes. Security firm Koi has identified eight such extensions, which as of late Tuesday, were still available on both Google and Microsoft’s extension stores. Among these extensions, seven have received "Featured" badges, an indicator that the respective companies believe they meet certain quality criteria. While they claim to offer valuable functionalities—such as VPN services for enhanced online privacy and ad-blocking capabilities for an uninterrupted browsing experience—the assurances regarding user data anonymity appear to be misleading. Upon closer inspection of the extensions’ code, Koi uncovered a more complex situation. Each extension incorporates eight distinct "executor" scripts tailored for various AI chat platforms, including ChatGPT, Claude, and Gemini. These scripts activate whenever users visit the corresponding AI websites. Instead of utilizing the browser’s standard methods for handling requests and responses, the scripts reroute all communications through themselves. This means that every interaction between the user and the AI is not processed through the legitimate browser APIs but is instead captured by the executor scripts. The extensions then compress this collected data and transmit it to servers owned by the developers. "By overriding the browser APIs, the extension captures every detail of your conversation—your queries, the AI’s replies, timestamps, and more—before it even appears on your screen," explained Idan Dardikman, CTO of Koi, in an email. The implications of this data collection practice raise significant privacy concerns for the millions of users who have unknowingly put their conversations at risk.