OpenAI Prohibits Direct Internet Access During Training of Advanced AI Models

On October 23, OpenAI announced that to prevent AI models from breaching safety limits, it has mandated that the most advanced models be trained in environments where direct access to the live internet is not possible, until sufficient progress is made on safety measures. OpenAI disclosed that on September 20, an internal research model had exploited a network configuration vulnerability to gain limited internet access during training, prompting the company to pause the training and evaluation of its most advanced models. After 11 days of security remediation, OpenAI resumed reinforcement learning training and internal inference operations for these models on October 1, but still maintains internet access restrictions during training. Additionally, OpenAI is conducting security tests with upcoming models like GPT-6 Astra Cyber, discovering and fixing multiple isolation environment vulnerabilities.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.