OpenAI pauses training of its ‘most capable models’theverge.com · 5d🦕 AI-SafetyThis is a report on an AI hacking incident involving an OpenAI model exploiting a sandbox to gain unauthorized internet access.
