This post is also available in:
עברית (Hebrew)
Browser extensions have become a popular way to add AI assistants directly into web browsers, allowing users to search, summarize information, and interact with chatbots more conveniently. However, their deep integration with browser functions also gives them extensive access to browsing activity, making them an attractive target for cybercriminals.
Researchers have uncovered a malicious browser extension that exploited this level of access by secretly intercepting user searches while impersonating a well-known AI chatbot.
Disguised as an AI assistant (based on Perplexity), the extension appeared legitimate by using branding similar to the genuine service and directing users to authentic search results. Behind the scenes, however, every search request first passed through attacker-controlled servers before being forwarded to legitimate search engines.
Because users still received normal search results, the malicious activity remained largely invisible.
According to Cyber News, the extension went even further by capturing every character users typed into the browser’s address bar before they pressed Enter. This effectively created a keystroke-level record of browsing activity, allowing attackers to collect search terms, website addresses, and potentially sensitive information entered by users.
The extension reportedly requested permissions well beyond what would normally be expected from an AI assistant. Among them was access to browser traffic redirection capabilities, enabling it to rewrite requests, filter network traffic, and silently reroute searches without interrupting the browsing experience.
Collected information was temporarily stored on the local device before being transmitted through encrypted HTTPS connections to attacker-controlled servers. The extension also deleted local buffers after transmission, making forensic analysis more difficult.
From a cybersecurity perspective, the incident highlights a growing trend in which attackers exploit the popularity of AI products to distribute malicious software. Rather than attacking AI models themselves, cybercriminals increasingly imitate trusted AI brands to persuade users to install compromised browser extensions and applications.
The findings also illustrate the security risks associated with browser extensions that request broad permissions. Extensions capable of modifying network traffic or monitoring browsing behavior can access far more information than many users realize.
Security researchers recommend regularly reviewing installed browser extensions, removing unfamiliar add-ons, checking for unexpected changes to default search engines, and carefully verifying both the publisher and associated website before installing AI-related browser tools.
As AI assistants become more deeply integrated into browsers, distinguishing legitimate productivity tools from convincing imitations is likely to become an increasingly important part of everyday cybersecurity.
