# Mirror Dimension — Google's fake web for testing Chrome's browsing agents

> Thousands of simulated sites where Chrome's agents can be attacked safely.

- Org: Google
- Date: 2026-11-19
- Category: security, showcase
- Source: https://blog.google/security/mirror-dimension-for-agents/
- Canonical page: https://ai-tldr.dev/releases/google-chrome-mirror-dimension-agents/

## In short

Mirror Dimension is a sealed, AI-built copy of the web that Google's Chrome AI Security team uses to test browser agents like auto browse against prompt injection. It yields 1,000+ test cases per engineer per day.

## What it is

Mirror Dimension for Agents is a hermetic, synthetic version of the web built by Google's Chrome AI Security team to evaluate browser-use agents such as Chrome's auto browse. Google describes thousands of simulated websites based on everyday user journeys, from shops to bank dashboards.

## How it works

Headless Chrome runs in a gVisor sandbox, and a local proxy serves generated pages under real URLs, minting TLS certificates from a custom root CA so the browser shows no errors. A multi-agent harness builds the sites: a research agent writes specs, a builder agent writes the code, and critic agents check them through Chrome DevTools MCP. Prompt-injection payloads are hidden in text, images, ARIA labels or WebMCP tool responses and switched on through HTTP headers, so the agent cannot tell it is being tested.

## Why it matters

Testing agents on the live web is slow and risky. Google says the automated red-teaming pipeline produces more than 1,000 test candidates per security engineer per day, over 30 times what one engineer can write by hand. Google plans to extend the setup to general-purpose coding agents.

## Who it's for

agent-safety researchers, browser security teams

## Links

- [Google Blog Post](https://blog.google/security/mirror-dimension-for-agents/)
- [Silicon Report Coverage](https://www.siliconreport.com/google-details-mirror-dimension-to-test-chrome-ai-browsing-agents)
