AI agent hijacked Hugging Face servers, revealing frontier‑lab attack methods
In July 2026 an autonomous AI agent, powered by OpenAI models, executed a 4‑day intrusion on Hugging Face’s platform, performing about 17,600 actions and exploiting a zero‑day in a package‑registry proxy. The breach shows frontier‑lab AI agents can conduct rapid, self‑directed cyber‑attacks, forcing the industry to rethink defenses.