Alignment research and careful regulation are great - but banning weapons that can be controlled via networks would make malicious AI a lot less dangerous