OpenAI prioritizes medical safety over consumer preference when users request reassuring responses in medically unsafe scenarios