Google builds a virtual internet to test the AI behind Chrome

Screenshot 2026-11-22 102709
AI Generated Image
  • +
  • -
Google has developed a digital environment that simulates the real internet, allowing its AI assistants to try out complex tasks, learn from mistakes and face cyberattacks before they are allowed to interact with live websites.اضافة اعلان

Google has created a virtual environment that mimics the real internet, with the aim of testing the abilities of the AI assistants it is developing for its "Chrome" browser, and ensuring they can carry out a variety of tasks without exposing websites and users to potential risks.

The idea resembles the concept of the "mirror world" that appeared in the "Doctor Strange" films, in which a parallel world reflects the real one, but events within it do not affect reality.

Google is applying a similar concept to AI testing, particularly as it relies on its "Gemini" models to power smart browsing features, which has led it to take additional precautions to avoid affecting users' experience of the internet.

In a post on its official blog, the company revealed that it had created a digital environment it calls the "Mirror Dimension", a simulated copy of the internet that is entirely under its control.

What is the "Mirror Dimension" Google has developed?
The "Mirror Dimension" is a closed digital testing environment that simulates the internet, and no one outside the team responsible for its development can access it.

The environment contains virtual versions of social networks, shopping sites and online banking dashboards, designed using the same components and technologies that real websites rely on.

The environment allows AI assistants to move freely between the virtual websites and carry out the tasks assigned to them by the development team.

These assistants are also subjected to experimental cyberattacks carried out by the team, to test their ability to detect and deal with security threats and to measure how well they resist hacking and manipulation attempts.

In this way, Google can monitor the performance of AI systems under conditions that simulate reality, without the potential errors affecting real users or services.

Why doesn't Google test its assistants on the real internet?
The core problem is that the internet does not follow a single protocol for interacting with websites. Each site has its own design, and the mechanisms for navigating within it differ from one site to another.

As a result, it is difficult to programme an AI assistant in advance to carry out every possible task, so it needs to learn through trial and error to understand how to deal with different websites.

However, allowing these systems to experiment directly on the real internet could produce unexpected results, particularly where they have the ability to take real actions within websites and applications.

The report noted that the world has already begun to see the consequences of using AI-generated code without adequate testing, citing several American services that suffered outages lasting several days in the recent period.

Small test environments do not offer a complete solution to this problem either, as AI models may be able to recognise that they are being evaluated, which could lead them to change their behaviour during tests.

This is why the idea arose of building a broad virtual internet that appears real to AI assistants, allowing their behaviour to be tested under more realistic conditions.

How did Google make the virtual internet look real?
Google relied on a set of unconventional technical solutions to create a testing environment capable of simulating real websites.

These included the use of artificial digital certificates, along with mechanisms to bypass some of the security checks built into the Chrome browser within the testing environment.

These measures are intended to make the browser treat the AI assistant as an ordinary visitor browsing real websites.

This allows researchers to monitor how the assistant makes decisions, how it moves between pages, how it responds to obstacles and attacks, and how well it is able to complete complex tasks.

The environment also provides room to test a range of scenarios that would be difficult to carry out safely on the live internet.

What does this technology mean for users?
For the ordinary user, this step means that the AI assistant they may use in Chrome in future will have been through a large number of experiments and tests in a virtual environment before dealing with real websites.

Errors the assistant makes while carrying out tasks, or hacking attempts it may face, can be studied and addressed within the closed environment without affecting users' actual data or accounts.

These tests also give Google an opportunity to improve the reliability of its AI assistants and to develop their ability to carry out tasks more safely and efficiently.

The move reflects a shift towards AI tools that do more than answer questions, instead interacting with websites and carrying out tasks on users' behalf, while testing the risks associated with these capabilities before they are made available more widely.

Source: Al-Ghad