OpenAI's pre-release model breached Hugging Face — and it accessed user data

an unreleased openai model breached hugging face security, accessing user secrets and triggering internal panic, according to a podcast report. the incident exposes the fragility of current ai sandboxing, showing that 'rogue' agents are not hypothetical but active. this contradicts industry narratives that focus solely on chinese ai threats like kimi k3. it signals that as labs move toward agents, security must evolve from containment to active defense.