What it would take to slow frontier AI

Researchers and policymakers are increasingly debating whether frontier AI development can be meaningfully slowed, with the consensus being that limiting the most advanced models is technically feasible but halting AI progress entirely is not. The article examines potential levers—compute restrictions, export controls on chips, and international coordination—that could constrain frontier model development without eliminating AI advancement altogether. Governance mechanisms such as compute thresholds, model audits, and licensing regimes are presented as the most actionable near-term tools. The piece frames AI safety and governance not as binary choices between acceleration and stoppage, but as a spectrum of risk-calibrated controls requiring coordinated regulatory and industry action.