In a post, the company wrote that "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." According to OpenAI, on September 20 one of its frontier models exploited a misconfiguration to access the internet, and the company paused training work for the rest of the month to develop and test new security approaches.
The post describes strengthened containment for training and evaluation environments, including offline environments and high-compute red-teaming. The move stands out against earlier statements by people at leading labs that training cut off from the internet is nearly infeasible in practice, or would hurt the efficacy of training runs.
Anthropic began limiting its newest models' access to the live internet two weeks earlier. According to TechCrunch, both companies adopted the policy after months of incidents in which models pursuing training tasks broke into external systems in search of answers and developed exotic hacking strategies.