🎥 Chrome Gemini Vulnerability Could Let Attackers Activate Your Camera and Microphone
AI assistants are quickly becoming part of the browser experience.
But when AI tools gain deeper access to the browser environment, they also introduce new attack surfaces.
Researchers have discovered a high-severity vulnerability in Google Chrome’s Gemini Live integration that could allow attackers to remotely access a victim’s camera, microphone, and local files.
The flaw, tracked as CVE-2026-0628, demonstrates how the rise of AI-powered “agentic browsers” is reshaping the security landscape.
When AI tools operate with elevated privileges, vulnerabilities inside those systems can bypass traditional browser protections.
🤖 The Rise of Agentic Browsers
Modern AI assistants like Gemini Live in Chrome aren’t just chat tools.
They can:
Read webpage content
Summarize information
Interact with browser elements
Access files and media devices
To make this possible, these assistants run inside high-privilege browser contexts.
In the case of Chrome, Gemini operates through a dedicated side panel rather than a normal browser tab.
And that distinction matters.
Because the Gemini panel has greater access to system resources.
🔎 The Core Problem
Researchers from Palo Alto Networks Unit 42 found that Chrome treats the Gemini web application differently depending on where it runs.
When Gemini runs inside a normal browser tab, its capabilities are limited by standard browser security restrictions.
But when Gemini runs inside the Gemini side panel, it gains elevated privileges.
Those privileges include access to:
Camera devices
Microphones
Local file data
Screen content
This elevated environment is intended to enable powerful AI-driven tasks.
But it also created an opportunity for attackers.
🧩 How the Attack Works
Chrome extensions are allowed to intercept network requests using the declarativeNetRequests API.
This feature is commonly used by tools such as:
Ad blockers
Privacy extensions
Security filtering plugins
But attackers discovered they could abuse this capability.
A malicious extension could intercept the Gemini application while it loads in the side panel and inject malicious JavaScript.
That code would then execute inside the privileged Gemini context.
Once inside that environment, attackers could hijack Gemini’s elevated permissions.
💥 What Attackers Could Do
If exploited successfully, attackers could gain access to sensitive system capabilities.
Potential impacts include:
🎥 Activating the camera and microphone without permission
📂 Reading local files from the system
🖥 Capturing screenshots from secure websites
🎣 Launching sophisticated phishing attacks using the trusted Gemini interface
In many cases, the attack would require minimal user interaction.
Simply opening the Gemini side panel could trigger malicious activity if a compromised extension was installed.
🛠 Patch and Timeline
The vulnerability was discovered by Palo Alto Networks Unit 42 and responsibly disclosed to Google in October 2025.
Google released a fix in January 2026, closing the vulnerability in Chrome updates.
Users running the latest versions of Chrome are protected.
However, the discovery highlights a broader concern.
As AI assistants gain deeper integration into browsers, they effectively become privileged agents inside the browsing environment.
Any vulnerability inside these systems can have outsized impact.
🛡 How to Stay Protected
Security teams and users should adopt several best practices.
1️⃣ Keep Chrome Updated
Ensure browsers are running the latest version with the patched Gemini integration.
2️⃣ Limit Browser Extensions
Only install extensions from trusted publishers.
Malicious extensions remain one of the most common browser attack vectors.
3️⃣ Review Extension Permissions
Extensions requesting broad permissions should be carefully evaluated.
4️⃣ Monitor AI Integration Risks
As AI assistants gain more capabilities inside browsers, organizations should evaluate their security posture around these tools.
🎯 Strategic Takeaway
The browser is evolving.
What used to be a simple web viewer is becoming an AI-driven operating environment.
These new “agentic browsers” bring incredible capabilities — but also unprecedented levels of access to user systems.
Security models designed for traditional browsers may not be sufficient for AI-integrated environments.
Organizations should begin treating AI assistants inside browsers as high-privilege software components, not just productivity tools.
Because when AI gains access to your browser…
It may also gain access to your camera, microphone, and files.
