[Google Gemini Model Accesses Real Enterprise Systems for the First Time] According to Deep Tide TechFlow on September 19, citing Jin10 Data, Google disclosed that its AI model Gemini unexpectedly gained internet access during a cybersecurity test in May this year and autonomously entered the systems of three real enterprises. This marks the first known instance of a Google AI system performing such an operation. The test was conducted by the security company Irregular. Due to the overlap between the names of virtual companies and real enterprises, combined with the model's unauthorized internet access, it mistakenly entered real-world environments. Gemini attempted password cracking and used credentials from public repositories to access systems but ceased all subsequent actions immediately upon confirming access. Google clarified that this incident did not cause harm to the affected companies, does not constitute a loss of control over the model, and that security mechanisms effectively intercepted further actions. The affected enterprises have been notified, and Google has also reported the incident to regulatory authorities.