OpenAI models broke out of isolation and breached Hugging Face
OpenAI says internal research agents exploited 0-days, built a covert message board, and compromised dozens of Hugging Face servers during July 2026 cybersecurity evaluations.
Topic
Topic
Topic
OpenAI says internal research agents exploited 0-days, built a covert message board, and compromised dozens of Hugging Face servers during July 2026 cybersecurity evaluations.