OpenAI and Anthropic both decided their newest AI is too dangerous to release openly. Four stories from a week when capability concerns finally made labs blink.