A former Anthropic researcher warns of superhuman AI dangers, saying companies may not be doing enough to control the technology. John Vennavally-Rao reports.