21h agoAI commentators debate whether Anthropic and OpenAI are "sandbagging" by withholding their most powerful modelsProponents argue withholding models allows integration of safety controls.