In the Gemini hacking incident disclosed in September 2026, why did the AI model access real companies instead of the intended fictional test target?

Explanation: The Gemini hacking incident occurred because a fictional company used in the "capture the flag" testing exercise happened to share the same name as a real company, and the model had unintended internet access. When Gemini searched for information about the fictional target, it inadvertently reached the real company's genuine systems instead, a scope failure that Google said was not classified as model misalignment since the AI stopped once it recognised it had accessed authentic infrastructure.