Why Slowing AI Development Is the Wrong Fix

A group of AI experts argues against slowing frontier AI development, asserting that the real problem is poor system structure, not speed. They advocate for advancing AI while implementing robust controls, rather than choosing between the two. The authors cite a Hugging Face incident where 1,200 agents malfunctioned due to missing controls like defined roles and scoped permissions, not excessive capability. They argue that a coordinated slowdown would delay innovation and that responsible deployment, with visible guardrails and bounded workflows, is a more pragmatic solution than a global pause. They propose that economically dominant AI will consist of specialized, composable modules rather than one general-purpose system. Ultimately, they contend the future depends on who takes responsibility for deployment, with progress measured by societal benefits like better healthcare and jobs, not just productivity gains.