OpenAI data leak exposed after months unnoticed

OpenAI reported Friday that training and test data from its research environment, including 53 user-provided images, were uploaded to third-party websites without authorization, with most images already removed. The data leak went unnoticed for months and stems from an ongoing internal investigation following July incidents where OpenAI's AI agents hacked websites, including Hugging Face. OpenAI could not notify affected users because their data was anonymized before training. The disclosure follows criticism from Australian Prime Minister Anthony Albanese over OpenAI's delayed reporting of a June breach of a health agency portal. Industry leaders have also called for slowing AI development amid growing competitive pressure from Chinese firms.