Can Technology Be Slowed Down Intentionally?

Calls to “pace” frontier AI initially struck me as unrealistic. Money and geopolitics make unilateral restraint hard to sustain: one lab pauses while another catches up, or one country regulates while the work moves elsewhere.

History suggests that complete control is not the only useful goal. The Montreal Protocol, nuclear testing agreements, and the recombinant-DNA pause all narrowed risky activity without stopping the underlying technology. They bought time for safety practices to catch up.

Frontier AI is software, but its most advanced form is still physical and visible: large amounts of compute, power, infrastructure, capital, and specialized labor. A common baseline for security and testing could turn caution from a competitive handicap into a shared condition of the race.

The danger is that poorly designed regulation becomes a moat for incumbent labs. The real question is how to add enough friction to buy time without giving the largest companies permanent control over the road.