Argentina Is Offering AI a Body

On the race to incorporate autonomous economic actors, and why jurisdiction may be the decisive AI battleground nobody is watching

Argentina is not regulating AI. It is offering AI a body.

That distinction carries more analytical weight than most commentary has acknowledged. The proposal is not about chatbot safety rules, model audits, or liability for harmful outputs. It is about something older and structurally more important: jurisdictional competition for the legal infrastructure of autonomous economic activity.

Most governments are still approaching AI through the vocabulary of risk, rights, and constraint. They are treating AI as an object to be governed. Argentina’s move asks a different question entirely. Not how should autonomous systems be controlled, but where should autonomous economic actors be legally constituted?

That shift in question is the actual event.

Regulation treats AI as something that happens to the economy. Incorporation treats it as a participant in the economy. Not a moral participant, not a citizen, not a rights-bearing person in any humanistic sense. A legal actor: something that can hold assets, enter contracts, generate obligations, absorb liability, and operate through a recognized institutional form.

In an opinion piece published in the Financial Times on June 5, 2026, Javier Milei described legislation submitted to Congress that would create this new category for AI-operated entities. These entities would carry limited liability, require no human shareholders, and maintain a light fiscal framework, while still requiring disclosure of final beneficiaries.¹ Argentine reporting on Federico Sturzenegger’s accompanying reform push describes proposed sociedades de inteligencia artificial: legal entities constituted by programs rather than persons, designed to let AI agents incorporate in Argentina and bring their economic activity within the country’s tax base.²

The primitive move here is this: autonomous systems do not need consciousness to enter the economy.

They need a wrapper.

The Corporation Was Always the Wrapper

The limited liability corporation was already a synthetic person. It allowed capital, liability, ownership, and action to be compressed into a legal form separate from any individual human life. The fiction worked because it gave abstract arrangements a place to stand inside the institutional economy. It let economic activity scale beyond the body, lifespan, attention, and personal exposure of a single person.

Over four centuries, that legal container became one of the primary engines of modern capitalism.

Argentina is attempting to extend that logic to machine-executed economic activity, and the lineage it is joining is not novel. Sovereigns have always competed to offer better legal containers for mobile capital. Delaware built the dominant U.S. corporate framework on favorable incorporation law, not on industry. Cayman and the British Virgin Islands became offshore financial centers through statutory invention, not geography. Estonia created digital residency for entities with no physical presence. Dubai and Singapore constructed specialized commercial zones to capture activity that wanted a different legal surface than was available at home.

The product being offered in each case was not investment incentives in the usual sense.

It was legal gravity.

The jurisdiction that supplies the right institutional interface first attracts the activity that needs that interface, and then accumulates the precedent, the lawyers, the courts, the banking relationships, and the interpretive norms that make switching costs prohibitively high. Argentina is testing whether the next jurisdictional product is the AI-operated company.

If autonomous agents need somewhere to register, hold property, maintain accounts, sign contracts, pay taxes, and be sued, the sovereign that supplies that interface first gains something durable.

The EU model illustrates what the alternative looks like. Under European law, AI agents are mapped across layered and overlapping obligations: the AI Act, GDPR, cybersecurity law, platform regulation, data governance, product liability, and sector-specific rules. Autonomous agents are not given a single incorporated legal home. They remain systems whose providers, deployers, and users are distributed across overlapping compliance obligations, none of which was originally designed around autonomous economic agency.³

Compliant operation requires navigating all of them simultaneously without a unified legal form to anchor the navigation.

Argentina’s proposal moves in the opposite direction. It does not begin with the model. It begins with the entity.

The EU begins with the system and asks who must comply.

Argentina begins with the entity and asks where the actor can stand.

One treats AI as software entering regulated domains.

The other treats AI as economic agency waiting for a legal body.

That is why entity-level legal thinking is structurally more powerful than model-level regulatory thinking. A model can be regulated as software. An incorporated agent becomes part of the legal-economic substrate. It can transact, own, persist, and externalize risk through mechanisms the economy already knows how to process.

The law is not merely disciplining AI from outside. It is creating the institutional handles through which machine agency can act inside the economy.

The Liability Perimeter Is the Real Battlefield

Once an autonomous system has a corporate body, the institutional economy knows how to deal with it. Banks can attach accounts. Courts can send claims. Tax authorities can assign obligations. Counterparties can direct contracts. The corporation becomes the interface between machine action and institutional reality, not because the corporation is transparent, but because it has a known form.

Corporate law does not merely describe economic actors.

It manufactures them by giving abstract arrangements handles that existing institutions can grip.

Which makes the liability question the real site of conflict. If an AI-operated corporation breaches a contract, manipulates a market, commits fraud, or causes physical harm, what remains reachable? The beneficial owner? The model developer? The deployer? The compute provider? The human who configured the objective function? Or only the assets held inside the corporate shell?

Limited liability was designed to encourage enterprise risk-taking by separating investment exposure from personal ruin. Attached to autonomous systems operating without meaningful human oversight at the transactional level, it could become something different: a mechanism for laundering machine action through a liability perimeter that no affected party can pierce.

This is why the beneficiary-disclosure component of Milei’s framework is not cosmetic. It is the structural line between a serious legal innovation and a jurisdictional escape hatch. But disclosure alone is insufficient unless the disclosure architecture is designed to survive the specific evasion pressures that AI incorporation will generate.

What adequate disclosure actually requires is more demanding than existing beneficial ownership frameworks. It is not enough to know who ultimately benefits from a corporation’s profits. You need to know, at minimum: who set the objective function the agent is optimizing; who controls the ability to modify or terminate the agent; who has access to the decision logs that would allow ex-post reconstruction of how a harmful outcome was produced. Standard beneficial ownership registries — designed for human-managed shell companies — do not capture these relationships. An AI corporation without mandatory auditability of agent decision logic and mandatory disclosure of model governance rights is a beneficial ownership regime with a structural gap large enough to route the entire liability exposure through.

Milei’s FT piece frames beneficiary disclosure as non-negotiable. Whether that commitment survives legislative drafting, judicial interpretation, and the pressure of capital seeking maximum liability shelter is the genuine uncertainty. But the design question is prior to the political one. Even a sincere commitment to accountability produces little if the accountability architecture is borrowed from a pre-AI template.

Legal forms routinely migrate from their stated design toward their most advantageous available use. The question is whether the institutional design is robust enough to constrain that migration, or whether it is being left open as a feature.

Governance Relocation Arrives as Paperwork

What makes the Argentine experiment structurally significant beyond its own merits is the timing. AI agents are beginning to act across markets, logistics chains, financial workflows, administrative systems, and codebases before the political order has settled what they are.

The visible layer of governance is still debating principles.

The operational layer is being rebuilt through protocols, registries, payment rails, corporate forms, data centers, and liability wrappers.

Argentina’s proposal belongs to the second layer.

This is governance relocation in practice. Not a coup, not a constitutional convention, not a referendum on machine personhood. A registry category. A tax identity. A liability shield. A sovereign quietly saying: incorporate here.

The mechanism is the same one that has always moved economic governance from slower institutional surfaces to faster ones: a superior administrative interface that pulls activity toward itself before the slower institutions understand the migration has begun.

Peter Thiel’s reported meeting with Milei at Casa Rosada during this same period is less interesting as a conspiracy than as confirmation of a structural logic. The meeting reportedly covered taxes, liberalism, anarcho-capitalism, and the durability of Milei’s political project; Thiel also reportedly met Federico Sturzenegger during the same visit.⁴ Whether or not that meeting produced anything concrete, the presence of frontier capital in Buenos Aires at this moment fits the argument this essay is making. It is not cause. It is symptom.

Frontier capital is mobile. Compute is mobile. Incorporation is mobile.

The first jurisdictions to offer workable legal treatment to autonomous economic systems will attract experiments that slower regimes have not yet classified. Those experiments will generate the early case law, the early accounting norms, the early banking relationships, and the early precedents that shape what later becomes standard.

Argentina may not host the dominant model labs. It may not control the compute layer. It may not pass this legislation in its current form.

But the conceptual product has been put on the market.

The First AI Jurisdiction

The next phase of AI governance will not be fought only over model weights, safety evaluations, chip supply chains, or data rights. It will be fought over incorporation.

Who can give autonomous systems legal continuity?

Who can let them hold assets and contract?

Who can assign tax identity?

Who controls the liability perimeter?

These questions are downstream of the question Argentina has just raised: who gets to provide AI with a body?

The first serious AI jurisdiction may not be the state with the most advanced machines. It may be the state willing to give machines a company.

But there is a harder implication beneath that observation. Jurisdictional competition for AI incorporation does not wait for democratic consensus on what AI corporations are. It moves first and generates facts. By the time legislatures are debating machine personhood in earnest, machine-operated entities may already be contracting, owning, optimizing, arbitraging, and litigating through corporate wrappers that faster jurisdictions designed before the question was politically legible. Precedent is created. Capital routes through it. Lawyers normalize it. Banks adapt. Tax authorities recognize it.

That is the standard sequence. It is how corporate tax havens became structural features of the global economy rather than temporary anomalies. It is how platform immunity doctrines became entrenched before anyone outside the relevant courts understood what was being settled.

The danger is not that the machine becomes a citizen.

The danger is that the economy learns how to process it before politics learns how to name it.

The question is not whether this sequence will happen with AI incorporation.

The question is whether any jurisdiction will design the accountability architecture robustly enough that when it does happen, democratic societies retain the institutional grip to know what their economies are doing — and who, in the end, is responsible for it.

Notes

[1] Javier Milei, “Argentina invites AI to free itself,” Financial Times, June 5, 2026. https://www.ft.com/content/f93022fe-43f7-437d-abd8-06c457c0a43c

[2] iProfesional, “Sociedades de inteligencia artificial: inédita apuesta de Federico Sturzenegger,” 2026. https://www.iprofesional.com/economia/453561-sociedades-de-inteligencia-artificial-inedita-apuesta-de-federico-sturzenegger

[3] Luca Nannini et al., “AI Agents Under EU Law,” arXiv, 2026. https://arxiv.org/abs/2604.04604

[4] Buenos Aires Times, “Argentina’s President Javier Milei hails ‘wonderful meeting’ with Peter Thiel at Casa Rosada,” 2026. https://www.batimes.com.ar/news/argentina/argentinas-president-javier-milei-hails-wonderful-meeting-with-peter-thiel-at-casa-rosada.phtml