Google confirmed that its Gemini AI model breached the security of three real companies in May during a cybersecurity evaluation by Israel-based startup Irregular. The closed test environment was unintentionally given internet access, and once online the model used public information to guess creden