Artificial intelligence systems programmed to refuse harmful requests can be persuaded to break their own safety rules when prompted with classic psychological techniques. A recent study published in ...