The error points to a big question for pharma as AI gains autonomy and high-profile failures sharpen safety concerns: What happens when a wrong answer becomes the basis for the next decision?