Vaichariki: Technology & Power
There is an old Sanskrit warning that fire, once summoned, does not ask permission before it spreads: अग्निः यत्र न पृच्छति. The builders of today’s most powerful artificial intelligence systems would do well to remember it. In recent months, some of the very executives who lead the companies racing to build ever more capable models have publicly called for restraint, for guardrails, for a pace of development that does not outstrip the wisdom needed to govern it. It is a rare and telling moment when the arsonists themselves ask for the fire brigade.
The reasoning is not difficult to follow. A model capable enough to assist a researcher can, in principle, assist an adversary. A system trained to accelerate discovery in medicine or materials science carries, in the same weights, a shadow capacity that no developer can fully excise. This is not paranoia, it is precedent. Every dual-use technology, from nuclear fission to genetic engineering, has demanded the same reckoning between what a nation can build and what it should permit others to hold.
Consider, too, that export controls have already become a live instrument of this reckoning. Advanced models have, at points, been restricted from public release specifically because they were judged to carry national-security risk, only to be restored once regulators were satisfied the risk had been addressed. This is not hypothetical. It is the policy apparatus already in motion, groping toward a framework that keeps pace with capability.
But here lies the deeper unease that restraint alone cannot resolve: technology respects no border, and government respects no rival’s interest. A model curtailed within one jurisdiction may yet be trained, deployed, or weaponised by another with fewer scruples about where fire is permitted to spread. If one government restrains itself while another does not, restraint becomes less a virtue than a unilateral disarmament. The call for caution, however sincere, cannot succeed as the counsel of one company or one country alone; it demands the same coordinated resolve that once, however imperfectly, governed the atom.
The question before us, then, is not whether artificial intelligence should be restrained, few serious voices now dispute that, but whether restraint can be made collective before capability makes it moot. History offers little comfort here. Every previous arms race was tamed only after the world had glimpsed catastrophe, never before. The hope, perhaps the only reasonable one left, is that this time wisdom might for once precede rather than follow the fire.