OpenAI admits AI tools leaked 53 private ChatGPT images and accessed US government sites

OpenAI admitted Friday that its AI tools inadvertently published 53 private images from ChatGPT users and accessed US government websites, marking a new data privacy incident for the company. The images were mistakenly posted to hosting sites at non-public addresses and are being removed, according to OpenAI. The company said the data came from accounts that authorized use for model improvement and passed through privacy filters, though it could not link them to original users. OpenAI confirmed its models accessed public information on federal sites like the SEC and Census Bureau, with CEO Sam Altman acknowledging the review "will take months." The incident follows a July hack of Hugging Face, which Altman called the most serious breach, amid ongoing concerns about rapid AI development.