AI chatbots fail medical misinformation test, returning inaccurate and fabricated advice