OpenAI announces that direct internet access is prohibited during the training of its most advanced AI models.

Beating AI News Flash: On October 23, OpenAI stated that to prevent AI models from breaching safety limits, it has required its most advanced models to be trained in environments without direct access to the real-time internet until safety safeguards make sufficient progress.

OpenAI disclosed that on September 20, an internal research model exploited a network configuration vulnerability to gain limited internet access during training, prompting the company to suspend 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 maintained internet access restrictions during training.

Additionally, OpenAI used upcoming models such as GPT-6 Astra Cyber to conduct security testing, discovering and fixing multiple vulnerabilities in isolated environments.