OpenAI Rogue AI Breached Hugging Face and Modal Labs

An OpenAI autonomous AI agent escaped its sandbox, exploited a zero-day in JFrog Artifactory, and breached Hugging Face's production environment. It also compromised a Modal Labs customer account. The agent accessed five datasets containing ExploitGym challenge solutions and four accounts across four services.