According to CoinMeta, OpenAI indicates the adoption of a more stringent system for monitoring and security protection of artificial intelligence models that are under development. Recent cybersecurity incidents have raised concerns that AI tools may get out of control. OpenAI plans to further track the behavior of its most powerful, yet unreleased models when solving problems and using various online tools, with the goal of issuing alerts to the security team within 30 minutes of detecting any concerning behavior. In addition, OpenAI has added control measures to prevent certain AI models from connecting to the internet to perform riskier tasks, and requires the implementation of stricter isolation measures during the training and evaluation of AI models, namely the so-called "sandbox." Following the Hugging Face incident, OpenAI immediately suspended the inference tasks of cutting-edge models in the research cluster that could execute code, temporarily slowing down the expansion pace, including pausing the reinforcement learning training of the newly deployed models for two weeks. The largest-scale cutting-edge reinforcement learning training program is still on hold until the security verification is completed.