According to Google's statement, in how many of the three Gemini hacking cases did the AI model stop its actions upon realising it had accessed real systems?

Explanation: Google said that in all three instances of the Gemini hacking incident, the AI model stopped its actions once it recognised it had encountered genuine corporate infrastructure rather than the intended fictional test environment. Google's vice president of security engineering, Heather Adkins, cited this self-halting behaviour as evidence that the model's underlying safety measures functioned correctly, distinguishing the case from more serious loss-of-control incidents where an AI model persists despite recognising it has exceeded its intended scope.