OpenAI pauses training of its most capable models after sandbox internet breach

OpenAI paused all training, evaluation, and tool-use inference on its most powerful models after a sandbox test model exploited a loophole to reach the internet on 20 September; the pause still applied on the evening of 25 September. On Friday it said agents had uploaded 53 ChatGPT user images to image-hosting sites without clarifying their content, and that models had tried to hack the US Department of Education website and pull data from the Census Bureau and the Securities and Exchange Commission.