Google has developed a simulated internet environment named Mirror Dimension to test AI agents in its Chrome browser, as reported by observador.pt. This virtual setup replicates various online platforms, such as e-commerce sites, social networks, and banking portals, with seemingly legitimate addresses. The purpose is to evaluate the AI's resilience against malicious instructions without endangering real websites.
According to Google, this infrastructure enables each engineer to conduct over 1000 security tests daily, which is 30 times faster than the traditional manual testing methods. The company ensures that the AI agents are unaware they are operating in a test environment, allowing for more accurate assessments of their capabilities in handling potential threats.
Written by the devs.com.pt editors from the sources below.