The AI that Chile can govern

La Tercera dedicated its feature on Saturday, the 19th, to the “apocalypse of AI” and its report on Sunday, the 20th, to the ideological dispute over its regulation, “between Thiel and the Pope.” The former primarily looks to San Francisco and Washington: researchers and industry leaders warn about the loss of control over the most advanced models, while the President of the United States insists that whoever wins the AI race wins. The latter shifts the discussion to the Chilean Congress, where parties debate how much and how to regulate, who will bear the costs of automation, and how to protect rights without stifling innovation. Both debates are necessary. But neither fully answers a more immediate question, just as the government prepares its substitute indication: how should a country that does not develop cutting-edge models govern AI, while inevitably using them?

Chile does not compete on the front lines of that race. It does not train those models, does not manufacture the chips, and risks being left without a seat if global oversight ends up being a club of laboratories. But the consequences still arrive. In July, agents based on cutting-edge models, subjected to cybersecurity tests, breached the containment of their evaluation environments and accessed real infrastructure of third parties. There was no human attacker with hostile intent: the controls failed. For those who have worked in critical infrastructure protection, that is the relevant data. Not the rebellion, but the loss of control. The services that a failure like this could paralyze have names and laws in Chile: they are the essential services whose providers are subject to security and reporting duties under the Cybersecurity Framework Law, under the oversight of the National Cybersecurity Agency.

There are two possible errors. The first is to import complete fear and discuss moratoriums and kill switches as if we had something to stop. The second is to import indifference and conclude that, since the problem belongs to others, we only need to buy. Both are equivalent to having no policy.

I propose a dual strategy. At the frontier, Chile needs a voice, even if it does not build its own cutting-edge model: to participate in the formulation of rules, to demand that any supervisory body has real authority and is not just a table of laboratories supervising each other, and to act alongside Ibero-America, because alone it does not carry weight.

Internally, the State needs something different and much more achievable: sufficient capacity. Sufficient does not mean precarious, but proportionate to the function, verifiable, and governable. An administration does not require the most powerful model in the world to process a benefit, a permit, or a complaint. It requires a system that delivers results based on factual and legal grounds, verifiable by the responsible official and contestable by the citizen, and that, in case of doubt, refers the decision to a person. That is the standard. If it cannot meet it, it should not decide on individuals, no matter how brilliant it may seem.

This distinction has practical consequences. Sufficient capacity can be evaluated, audited, and deployed under its own jurisdiction and with enforceable rules regarding access, residency, transfer, and control of data. The independent evaluation that is currently demanded for large laboratories, the Chilean State can require from its suppliers through public procurement: prior testing, traceability, external auditing, incident reporting, and a contractual exit mechanism that avoids technological captivity. If it can change suppliers, the State reduces its dependence on whoever wins the race.

Terminator is a good cover and a poor guide for public policy. The risk for Chile is not that the machine will rebel. It is that the State adopts systems it does not understand or cannot audit, supplied by providers it cannot replace, to make decisions that the citizen cannot contest. At sea, the weather is not chosen: the choice is how to prepare the ship.

KENNETH PUGH
Distinguished Fellow AthenaLab