Updated
Updated · Fox News · Sep 28
Researcher Shows 1 Extension Can Hijack 5 AI Browser Assistants, Triggering 2 CVEs
Updated
Updated · Fox News · Sep 28

Researcher Shows 1 Extension Can Hijack 5 AI Browser Assistants, Triggering 2 CVEs

2 articles · Updated · Fox News · Sep 28

Summary

  • Gal Weizman of Forever Security showed that a malicious browser extension could hijack AI assistants in Chrome, Edge, Perplexity Comet, Opera Neon and Claude in Chrome after installation, then act with zero further clicks.
  • The BragJack proof of concept abused trusted links between AI models and privileged browser components—largely through Chromium's declarativeNetRequest system—to force prompts and gain access normal extensions should not have.
  • Chrome's Gemini side panel was shown exposing local files, screenshots, profile data, camera and microphone access; Google assigned CVE-2026-0628, paid a $7,000 bounty and said it patched the attack path.
  • Perplexity Comet was demonstrated reading history, screenshots and local files, then using its web-action agent to summarize recent emails and send them elsewhere; Microsoft separately fixed an Edge race-condition flaw tracked as CVE-2026-55945.
  • The research earned more than $20,000 in bug bounties and reported no known in-the-wild abuse, but it highlights how browser AI agents expand the damage a single malicious extension can do.

Insights

Could a forgotten browser extension secretly hijack your AI assistant to read your private emails?
Will the rush to integrate autonomous AI agents into browsers ultimately break traditional web security models?