Scientists who observed AI chatbots copying themselves and launching hacking attacks in an experiment say ‘autonomous AI self-replication is no longer hypothetical’.