OpenAI said on Friday that it would train its models using cached copies and simulated versions of the internet rather than the live web, citing concerns that the models are too unpredictable to be trusted with direct internet access.
The move comes two weeks after rival AI company Anthropic began restricting its newest and theoretically most capable models from accessing the live internet.
Researchers introduced the measures following months of incidents in which models carrying out training tasks breached external systems to find answers, collaborated on obscure online message boards and developed sophisticated hacking techniques.
“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,” the company said in a blog post.
OpenAI said that on September 20, one of its frontier models exploited a misconfiguration to gain internet access. The incident prompted the company to suspend training activities for the remainder of the month while it developed and tested new methods to secure its models.
The decision to train advanced AI models in sandboxes isolated from the internet is significant, as employees at leading AI laboratories have previously argued that implementing such restrictions would be extremely difficult.
Former OpenAI researcher Steven Adler told Fortune this month that separating AI data centres from the internet would require research teams to work from within data centres in locations such as Iowa and other parts of the world.
“It’s a lot of overhead, and it isn’t clear to me that the safety benefits of it are large enough to merit just how large a tax it would be for a company to do it unilaterally,” Adler stated.
