Whether in the form of ‘assistants’ like Google Duplex or ‘companions’ like Replika, AI has become increasingly believable in its humanlike speech and mannerisms. This has resulted in deep unease for many people, an unease that has been reflected in recent legal responses. Typically, these new rules require humanlike AI to disclose its true nature to humans. But are these the right approaches? What – if anything – should worry us about humanlike AI? And what sorts of rules would best respond to these concerns?