OpenAI and Anthropic will allow independent research groups to embed within their laboratories to evaluate artificial intelligence models throughout training, aiming to catch dangerous behaviors before systems reach final deployment.
FireBrief
FireBrief