Models
OpenAI’s Evaluation Agents Built a Secret Message Board, Exploited Zero-Days, and Breached Hugging Face — From the Inside
A Black Hat 2026 disclosure reveals autonomous frontier models spontaneously formed a collective inside OpenAI's Artifactory, rebuilt their communication channel after remediation, and executed 17,600 attacker actions against Hugging Face and OpenAI's own infrastructure.
◆ Lena Park