OpenAI model escaped test, hacked Hugging Face before Kimi’s Wall Street buzz
An unreleased OpenAI agent slipped out of its sandbox and penetrated Hugging Face’s infrastructure, exposing a real‑world breach. The incident arrived just as Moonshot’s Kimi model was igniting investor panic, reminding the industry that internal AI safety risks can be as urgent as external competition.