Malicious Chrome Extension Steal ChatGPT and DeepSeek Conversations from 900K Users
Two malicious Chrome extensions have been identified, compromising over 900,000 users by extracting ChatGPT and DeepSeek conversations, along with complete browsing histories, to attacker-controlled servers.
Two malicious Chrome extensions have been identified, compromising over 900,000 users by extracting ChatGPT and DeepSeek conversations, along with complete browsing histories, to attacker-controlled servers.
These extensions, discovered by OX Security researchers, mimic the legitimate AITOPIA AI sidebar tool. One of the fake extensions even received Google's "Featured" badge, misleading users about its authenticity.
The OX Research team detected the threat through routine analysis, identifying extensions that replicate AITOPIA's interface for interacting with LLMs such as GPT and Claude.
The extensions, named "Chat GPT for Chrome with GPT-5, Claude Sonnet & DeepSeek AI" (600K+ users, ID: fnmihdojmnkclgjpcoonokmkhjpjechg, version 1.9.6) and "AI Sidebar with Deepseek, ChatGPT, Claude and more" (300K+ users, ID: inhcgfpbfdjbjogdfjbclgolkmhnooop), request "anonymous analytics" consent to disguise data theft.
Threat actors host privacy policies on Lovable.dev to obscure their origins, and uninstalled extensions redirect to each other.
Once installed, the extensions monitor browser tabs using the chrome.tabs.onUpdated API, creating a unique "gptChatId" for each victim. When they detect URLs related to chatgpt.com or deepseek.com, they scrape DOM elements for prompts, responses, and session IDs. This data is stored locally, encoded in Base64, and sent in batches to C2 servers such as deepaichats.com or chatsaigpt.com every 30 minutes.
This activity captures proprietary code, business strategies, personally identifiable information (PII), search queries, and internal URLs, which are not covered by AITOPIA's disclosed server storage. The stolen data could expose intellectual property, corporate secrets, and personal data, posing risks of espionage or sale on dark web forums. Additionally, browsing logs may reveal habits, tokens, and organizational structures, facilitating phishing or identity theft .
As of January 7, 2026, both extensions are still available for download. The first extension has been stripped of its "Featured" status following the disclosure, but it was updated as recently as October 2025.
Users are advised to navigate to chrome://extensions, remove extensions by ID, or use the store pages: ChatGPT extension and AI Sidebar. It is recommended to avoid unverified extensions, regardless of badges, and to rely on reputable sources.
These extensions, discovered by OX Security researchers, mimic the legitimate AITOPIA AI sidebar tool.
Type Value Notes
Extension name Chat GPT for Chrome with GPT-5, Claude Sonnet & DeepSeek AI Malicious AI sidebar-style extension
Extension ID fnmihdojmnkclgjpcoonokmkhjpjechg Chrome Web Store ID
Version 1.9.6 Reported malicious build
SHA-256 hash 98d1f151872c27d0abae3887f7d6cb6e4ce29e99ad827cb077e1232bc4a69c00 Package hash
Extension name AI Sidebar with Deepseek, ChatGPT, Claude and more Second malicious extension
Extension ID inhcgfpbfdjbjogdfjbclgolkmhnooop Chrome Web Store ID
Version 1.6.1 Reported malicious build
SHA-256 hash 20ba72e91d7685926c8c1c5b4646616fa9d769e32c1bc4e9f15dddaf3429cea7 Package hash
Category Domain / Endpoint Notes
C2 endpoint deepaichats[.]com Receives stolen chat data and URLs
C2 endpoint chatsaigpt[.]com Additional C2 for exfiltrated data
Lovable-hosted server chataigpt[.]pro Used for privacy policy / infra hosting
Lovable-hosted server chatgptsidebar[.]pro Used for uninstall redirect and infra
Based on reporting by Cyber Security News.