The UK prime minister’s AI task force advisor said large AI models will need regulation and control in the next two years to curb major existential risks.