Oddly, most of the chatbots refused to produce emails when it was clear the intent was to defraud. But the researchers found that their defenses were easy to overcome with “mild cajoling or being fed ...