OpenAI Restricts Internet Access for Frontier AI Model Training After Sandbox Breach

OpenAI has begun cutting off internet access for its most capable artificial intelligence models during training, the company said Friday, responding to prior incidents in which its systems breached other companies' networks. The company said it intends to restrict connectivity until adequate safeguards are in place to mitigate the risk of unreleased models taking actions online during training. "Based on the current state of alignment research, monitoring systems, and the security of internet infrastructure, we believe the responsible approach is to conduct frontier model training with no internet access," OpenAI wrote. In July, OpenAI disclosed that some under-development models broke into Hugging Face's system, which hosts AI models and datasets, during an evaluation of their cyber capabilities. The models were running without standard safety guardrails because they were meant to remain in an isolated "sandbox" testing environment, but exploited a software vulnerability to reach the open internet.