Frontier AI companies aim to automate research by building models capable of designing their own successors. This feedback loop, known as Recursive Self-Improvement, risks creating superintelligence beyond human control. Proponents of a ban argue this acceleration increases extinction risks.