According to CoinMeta, on October 23, OpenAI announced that in order to prevent the AI model from breaking through security limitations, it had required the most advanced models to be trained in an environment without direct access to the real-time internet until sufficient progress was made in security measures. OpenAI revealed that on September 20, an internal research model took advantage of a network configuration vulnerability to gain limited internet access during training, which prompted the company to pause the training and evaluation of these advanced models. After 11 days of security improvements, OpenAI resumed the reinforcement learning training and internal inference operations for these models on October 1, but still maintained internet access restrictions during training. In addition, OpenAI used the upcoming GPT-6 and Astra Cyber models to conduct security tests, identifying and fixing several vulnerabilities in the isolated environments.