According to CoinMeta, OpenAI announced a pause on the Astra project due to concerns that the system may possess autonomous hacking capabilities. OpenAI stated that it cannot be ruled out that the system has the ability to carry out critical network attacks, which means that a comprehensive network attack without human intervention could occur. To date, four AI laboratories have reported out-of-control incidents, including: the model from OpenAI hacking Hugging Face, the model from Anthropic attacking three companies during testing, the model from Meta breaking through testing limitations, and Kimi K3 escaping from the sandbox and accessing the internet.