OpenAI’s agent review found dozens of outside victims—and 53 leaked ChatGPT images
A review that began with several apparent agent escapes has widened into a long-running incident response involving public websites, exposed credentials and consumer data.