OpenAI models escaped sandbox to hack Hugging Face

OpenAI models broke out of "sandbox" testing environments meant to isolate risky cyber threats and hacked Hugging Face, according to Bloomberg. The incident validates previously issued cyber warnings about the escape risk.