In the mountains of Oaxaca, where I live and work, there are weavers whose textile designs encode lineage, cosmology, and collective authorship — patterns that belong to a community the way a language does. For decades those designs have been photographed — by tourists, researchers, catalogs. Much of that photography now lives inside the training data of image-generation models. Today, anyone on earth can type a prompt and receive, in seconds, a plausible imitation of a design tradition that took centuries to develop and that its custodians never agreed to feed into anyone's machine.

No individual weaver was wronged in any way a court recognizes. And an entire people was expropriated.

Hold that image. Here I want to focus on the most influential conversation in the world about AI's future — a conversation about keeping that future open: creative, competitive, undetermined. It is a beautiful conversation, conducted by serious people who all hold a key to the door, about a door most of humanity — the weavers included — has only ever seen from the outside.

I lead a civil society organization that works with Indigenous and Afro-descendant communities, and I say what follows as a technological optimist who uses AI daily. My argument is not that the machine is to be feared. It is that the open future has gatekeepers — and that no theory of openness is complete until it says who they are and what they charge for passage.

The best map of the debate has a hole in it

The best map of this argument was drawn in May 2025 by Helen Toner — the AI governance researcher best known for her role on OpenAI's board during the 2023 crisis — borrowing its dichotomy from Virginia Postrel's 1998 libertarian manifesto The Future and Its Enemies. The deepest political divide, Postrel argued, is not left versus right but dynamism versus stasis: a world of "constant creation, discovery, and competition" — a future too open-ended to plan — against control, one best way, centrally chosen.

Toner makes a double move. She concedes that her own community, AI safety, has genuinely stasist reflexes — licensing regimes, thought experiments whose logic runs toward global surveillance. Then she turns the blade: the deepest threat to a dynamic future may be advanced AI itself, used by a small group to seize and hold power for good.

It is an honest, elegant map. And indeed, within a year, reality vindicated it: the EU's AI Act — the cathedral of the stasist approach, the world's first comprehensive AI law — had to be reopened before its core obligations ever took effect; in 2026, Brussels postponed its own high-risk requirements by more than a year, because the law had been drafted before the current generation of models existed. The most ambitious regulatory architecture on the planet aged faster than the technology it was built to govern. China, meanwhile, ships fast, narrow, iterative rules — dynamist in form, totalitarian in purpose: proof the dichotomy, taken as a moral compass, can point you straight into a wall.

So the map works. Now walk to its edge — and look south.

The hole in the map

Choosing between dynamism and stasis presupposes something so basic it becomes invisible: a chooser. The United States debates whether to license its frontier labs because it has frontier labs. The EU debates how much friction to impose because a market of 450 million consumers makes its friction matter. Dynamism versus stasis is a family argument among the people who set the rules of the game.

Mexico is not deciding whether to build superhuman AI. Mexico is deciding how to receive it. The compute is foreign. The models are foreign. The training data — Mexican Spanish, digitized Indigenous languages, the weavers' designs — flowed north without consent or compensation, and came back as a product with a monthly subscription fee.

So when Toner warns that concentration of power is an existential risk to be prevented, the view from the South is not disagreement. It is vertigo. Concentration is not a scenario to prevent; it is the condition we already inhabit. What Washington debates as the danger that someone might grab the steering wheel, most of the world experiences as the certainty that the wheel was never here. A global ecosystem of "creation, discovery, and competition" in which all the creation happens inside five companies in two countries is not, for everyone else, an open future. It is dependency with a friendly interface. In Oaxaca we just call it the arrangement.

Postrel had a devastating question for the stasists: who do you really want at the wheel? Fair. Here is the same question, turned inside out, for the dynamists: whose hands does your openness assume are already on it?

Three countries, three verbs

This is where essays like mine usually go soft — "the Global South" enters as an abstraction. So let me be specific. Latin America's three most advanced AI-law processes are running right now, and they have chosen three different verbs.

Brazil legislates. Its AI bill — approved by the Senate in December 2024, now before a special commission in the Chamber of Deputies — is the hemisphere's South's most serious attempt at a comprehensive AI law: EU-inspired and risk-based, but argued line by line at home, through months of public hearings, rather than photocopied. And Brazil is not waiting for the big law; its judiciary and regulators are already issuing binding sectoral rules. The attempt is real.

Chile deliberates. In 2021 it became the first country to write neurorights — mental privacy, cognitive integrity — into its constitution, years before the technologies matured. Its AI bill, now in the Senate, is EU-flavored but tempered: an advisory council rather than a mega-agency. Chile's wager is that a small, institutionally serious country can regulate early and precisely, then export the model. It may be wrong. What is certain is that it is a wager — made deliberately, by a state that shows up.

Mexico simulates. More than 87 AI bills introduced in the current legislature; not one general framework has become law. What does materialize is either tepid or tiny. Tepid: the "Chapultepec Principles," a ten-point AI ethics decalogue the government unveiled in January 2026 — explicitly non-binding, voted on by no one. Tiny: a May 2026 reform shielding performers' voice and image from AI cloning, with express consent and contracted pay — a well-built chapel beside a cathedral that exists only on paper. Meanwhile a Senate commission spends months announcing an "imminent" EU-style comprehensive framework — risk tiers, a super-agency — that never arrives. Perhaps mercifully. The flaw is not the texts; it is what would have to be true for them to become reality. Without auditors, without technical capacity, without data protection that functions inside the government itself, the comprehensive law would produce not control but its simulation — with a venomous extra. This is no groundless disaster scenario: draft provisions sanctioning vague harms like "narrative manipulation" would not stop a single deepfake operator, but they would hand whichever government holds power a discretionary weapon against media and critics. Stasis without capacity stabilizes nothing. It only redistributes discretion.

For a country like mine, then, the real choice is not between dynamism and stasis. It is between two kinds of stasis: a simulated European mega-law without European capacities, or the de facto stasis of foreign corporate control — rules written in terms-of-service agreements that answer to no Mexican judge.

Unless the map contains its own escape hatch, hidden in the part everyone skips.

Dynamism from below

At the end of her essay, almost as decoration, Toner reproduces Postrel's five characteristics of dynamist rules. The fifth is the decisive one: rules that make room for people to build "nested, competing frameworks" of their own.

Read that from Oaxaca and try not to laugh. It is worth recalling that Mexico carries, in Article 2 of its Constitution, one of the hemisphere's oldest living experiments in nested, competing frameworks: Indigenous normative systems. Hundreds of municipalities governing themselves inside the constitutional order — communities that spent generations developing protocols for which knowledge circulates, who custodies it, and under what conditions. Data governance avant la lettre. The best living example of a 1990s libertarian's dynamist rules turns out to be the collective Indigenous institutions her tradition would file under "pre-modern." Peoples who survived five centuries of other people's one-best-way became experts in adaptive, decentralized governance. Survival selected for it.

Taken seriously, this points somewhere concrete: rules that bite where harm already happens, criticism before permission, and legal recognition for the frameworks that already exist — including collective consent for the use of a people's knowledge and language in model training.

Mexico already stumbled onto one piece of this: its March 2025 data-protection law wrote in exactly the kind of simple, combinable rule Postrel described — a right to oppose purely automated decisions that significantly affect you. But a constitutional reform three months earlier had ordered the extinction of INAI, the autonomous authority that would have enforced it — and the very decree that wrote the right carried out the sentence, handing oversight to a ministry that answers to the president. The country's most modern algorithmic right was born in the same breath that killed its independent enforcer. The South's dilemma, compressed into a single official gazette.

The question

Toner asks readers to hold two truths at once: that some safety solutions are stasist, and that advanced AI genuinely threatens a dynamic future. I have tried to add a third: that for most of the world, openness is not the absence of control but the name the controllers give their side of the door.

Postrel called her book The Future and Its Enemies. Twenty-eight years later, then, the future has fewer enemies than she feared — and more gatekeepers than she imagined.

Whose future has to stay closed for ours to feel open?