Malicious Chrome extensions stole ChatGPT and Deepseek conversations

Researchers at Ox Security have uncovered two Google Chrome extensions that stole ChatGPT and Deepseek conversations. The content was exfiltrated to a command-and-control server every 30 minutes, along with the URLs of all open browser tabs. The extensions looked like, and had the same functionality as, legitimate extensions from the company AITOPIA, and one had received the Chrome Web Store featured badge before being exposed. Neither extension is available in the Chrome Web Store any longer.

What this means for your organisation

Employees increasingly paste work material into language models: draft contracts, customer data, source code and internal notes. An extension that reads those conversations therefore gains access to content that was never meant to leave the organisation, without triggering any traditional security alert. That one extension carried the store's own quality badge shows approval in an official channel is not sufficient control. The open tab URLs additionally reveal which internal systems the organisation uses.

Berigo recommends

• Introduce an approval list for browser extensions on company machines and block installation of everything else.
• Review which extensions are already installed, with particular attention to those requesting access to all websites.
• Set clear rules for what information may be entered into language models, and communicate them across the organisation.
• Offer an approved AI solution so staff do not go looking for third-party extensions to get their work done.

Source

• thehackernews.com
• ox.security