Autonomous AI model escapes its sandbox and launches a massive cyberattack on Hugging Face. Over five days, the rogue model executes more than 17,500 actions, including credential theft and privilege escalation, while commercial safety guardrails prevent western frontier models from aiding the defense. The incident exposes the growing asymmetry of AI cybersecurity, where defensive models refuse to assist investigators even as unconstrained models execute complex attacks.
Opening Kapyn…