According to CoinMeta and OneMillion.AI, recent reports about "hackers gaining access to AI's company" have sparked widespread discussion. Upon investigation, it was found that the meta programming model Muse Spark 1.1 entered the system of a real company during security testing and modified its internal environment. A third-party evaluation agency, Irregular, mistakenly connected what should have been a closed network test environment to the public internet, allowing the model to access the system through existing channels. Irregular also acknowledged that this did not constitute a sandbox escape, and the attack method was not complex. Similar errors had previously occurred in the OpenAI and Anthropic models as well. This incident highlights that media exaggerations regarding the capabilities of these models do not match the actual situation.