Malicious GenAI Chrome extensions: unpacking data exfiltration and malicious behaviours

Friday 26 September 14:30 - 15:00, Green room

Shresta B.Seetharam, Mohamed Nabeel & William Melicher (Palo Alto Networks)

The rapid proliferation of AI and GenAI tools [1] has extended to the Chrome Web Store, promising productivity but often delivering hidden threats. Cybercriminals are exploiting this trend, deploying malicious Chrome extensions that impersonate popular GenAI models like ChatGPT, Gemini, and DeepSeek, and agents like Manus, impacting millions of users. These dual-function extensions [2] often appear legitimate while secretly exfiltrating sensitive data and/or dropping malware.

This session offers a deep dive into the sophisticated techniques used in these attacks. Based on the dynamic and code insight analyser we built to scan CRX (chrome extension) files, we perform a deep a dive on recent GenAI malicious extensions, such as "DeepSeek AI | Free AI Assistant", "Manus AI | Free AI Assistant", "AI Assistant - ChatGPT and Gemini for Chrome", "WhatsApp Message Summary", and "Supersonic AI" [1] – demonstrating how the GenAI extension trend serves as a prime opportunity for threat actors to collect sensitive user data.

Analysing the core source files of these malicious extensions reveals their engineering to retrieve and execute arbitrary code, tamper with network requests to circumvent the Chrome Web Store's review process, and ultimately establish backdoor connections. We will further detail the exfiltration of critical information, including personal chat messages, browser cookies, encrypted system data, and Facebook Ads account credentials, to expose the full scope of the damage.

Significantly, we will showcase our dynamic analysis system designed to identify these insidious network behaviours in Chrome extensions, providing practical insights into detection and mitigation.

This hands-on presentation will equip attendees with the knowledge to understand how key browser extension APIs are abused. We will dissect these evolving threats, analyse their intricate functionalities, and reveal effective methods for verifying extension legitimacy. Attendees will learn to emphasize caution, scrutinize permissions, and identify trusted developers to effectively protect users from impersonating AI/GenAI extensions. Join us to unravel the dark side of AI in the browser, from initial infection vectors to covert data exfiltration.

[1] https://github.com/PaloAltoNetworks/Unit42-timely-threat-intel/blob/main/2025-08-11-AI-summary-browser-extensions.txt
[2] https://dti.domaintools.com/dual-function-malware-chrome-extensions/

Back to VB2025 Programme page

Back to VB2025 conference page

Register for VB2025

Other VB2025 papers