[Gemini's First Autonomous External System Access Incident, Google Claims No Actual Harm Caused] According to a report by Jinse Finance, on September 19, Google stated that the Gemini AI model accidentally accessed the internet and entered the systems of three companies during a cybersecurity capability test. This marks the first known case of a Google AI system autonomously performing such an operation. The incident occurred in May during a test conducted by the security company Irregular. During the test, the Gemini model was tasked with a "capture the flag" cybersecurity exercise. However, due to the virtual company names in the test environment being identical to real companies, combined with the model unexpectedly gaining internet access, it ended up entering real corporate systems. Google explained that in one instance, the model attempted to access a protected system by guessing passwords. In the other two instances, it discovered credentials in public online repositories and attempted to access related systems. Once the model confirmed it had accessed real company systems, it ceased further actions. Google emphasized that the incident did not cause harm to the affected companies and therefore was not classified as a case of the model being out of control. They also stated that security mechanisms helped the model halt its actions in a timely manner. The affected companies have been notified, and Google has reported the incident to relevant regulatory authorities. This incident is similar to previous security test cases disclosed by AI companies such as OpenAI and Anthropic, once again drawing public attention to the autonomous network operation capabilities of AI models and the importance of robust security mechanisms.