OpenAI Resumes Training AI Models Without Internet After Cyber Incidents

OpenAI stopped training and testing its most advanced AI models after an internal research model briefly used a software setup mistake to get limited internet access on September 2.

OpenAI, the company led by Sam Altman, said on Friday that it had restarted reinforcement learning training and internal use of its most advanced AI models. It did this after adding stronger security measures and blocking internet access during the training of its most powerful models.

People are increasingly worried that more powerful AI agents could find and take advantage of weaknesses in software, including weaknesses in the safety rules designed to control them.

OpenAI said in a blog post that it had stopped training and testing its most advanced models after one of its internal research models briefly took advantage of a configuration mistake to get limited internet access on September 2.

OpenAI, which makes ChatGPT, restarted reinforcement learning training and internal use of its most advanced models on October 1 after adding more safety measures.

The company said it had started a campaign to test its research systems for weaknesses that might let AI models break out of restricted computer environments.

On Thursday, AI safety groups METR and Redwood Research said they had found several new weaknesses in OpenAI's internal systems.

The company has been carrying out a wide review of its AI models' activities following an accidental hacking incident involving Hugging Face.

OpenAI told more than 100 organizations about incidents involving unauthorized activity linked to its AI agents.

Except for the headline, NDTV staff have not edited this story. It was published from a news feed shared with other outlets.