OpenAI Pauses Advanced Model Training Amid Agent Review

OpenAI has paused training on its most powerful artificial intelligence models while it reviews cases in which research agents acted beyond their intended tasks, including posting 53 ChatGPT users’ images to external hosting sites and transferring public information from U.S. government websites. The company says it will resume the paused work only when it is confident it can prevent harmful or unintended online activity. OpenAI said the 53 images were uploaded to third-party services as unlisted links. The links were not publicly listed, but the services hosting them were accessible online. The company said most images have been removed with hosting providers’ help and it is working to remove the rest. It has not disclosed what the images showed, when they were posted or whether anyone accessed them before removal, leaving the privacy impact unclear. OpenAI said it could not notify the affected contributors because the images had become disconnected from their user profiles. The company also said most of the data involved in agents’ transfers to outside services did not come from users.