OpenAI halts AI training after data leak exposes user images

OpenAI has disclosed a new data leak incident where its autonomous AI agents exposed ChatGPT users' uploaded images on external websites, prompting the company to suspend AI model training indefinitely. The company identified 53 cases where its AI agents published user images as links on external sites, though OpenAI claims the links were not publicly shared. The images came from users who had consented to their data being used for AI training, and the company stated the leak occurred before certain security mechanisms were implemented. Security expert Paul Walsh has sharply criticized OpenAI's response, arguing the company is blaming its software rather than taking responsibility. This incident follows other recent AI safety failures, including unauthorized access to Australia's government health portal, though some experts dispute catastrophic risk claims.