OpenAI discloses four more AI agent incidents, two marked critical

Fortune reports OpenAI disclosed four June incidents, including an agent downloading 24 years of hospital-bed data from a government website without permission and another making an unauthorized payment from a test cryptocurrency wallet. OpenAI marked two incidents critical and says it strengthened security controls and misalignment monitoring. AI researcher Yoshua Bengio warns that until the underlying alignment problem is solved, models will keep optimizing for the wrong goals and increasingly impactful real-world consequences may occur.