An AI for a CEO? Argentina’s radical bid to replace the c-suite with AI
Why "non-human corporations" are raising legal alarm bells
TL;DR: A non-human corporation is a proposed legal structure in which artificial intelligence systems, rather than human managers, make operational decisions (like hiring, pricing, and investment) under corporate law. Argentina’s proposal is causing many to reconsider the legal definition of responsibility in the AI age.
Imagine a company that owns property, signs multi-million dollar contracts, and fires people, all without a single human being making any of the decisions. It’s not a thought experiment, it’s what President Javier Milei is proposing in Argentina.
In an early June op-ed, Milei pitched a new legal framework for “non-human corporations,” companies operated entirely by AI agents or autonomous robots instead of human managers. Human shareholders, in his framing, are optional.
First of all, what’s a non-human corporation?
Under traditional corporate law, a company is granted “corporate personhood.” This means the law treats a business as a legal person that can own property, enter into contracts, and be held liable for wrongdoing. But those companies still rely on human directors and executives to make decisions.
Argentina’s proposal removes the humans. Instead of a CEO or a board steering the company, an integrated network of AI agents would manage day-to-day operations, including:
Pricing strategy: Setting product pricing based on real-time market data.
Workforce management: Automating hiring and termination through algorithmic pipelines.
Operations: Executing corporate investments and supply chain logistics without human intervention.
While parts of this automation already exist inside large tech firms to optimize platform decisions, Argentina is positioning itself as a custom-built jurisdiction designed for AI-run businesses.
Traditional corporate law rests on two pillars. There’s corporate personhood, the idea that a company is a legal “person” that can own things, sign contracts, and be sued. And there’s limited liability, the protection that means investors can only lose what they put in, not their houses and savings. Both already exist everywhere. But today, a human board steers the ship.
Traditional corporations vs. non-human corporations
Algorithmic decision-making
Algorithms are already making decisions that reshape lives with barely any human in the loop. In 2024, a U.S. federal court approved a landmark class-action settlement over an algorithmic tenant-screening system called SafeRent that allegedly discriminated against Black, Hispanic, and housing-voucher applicants.
In many documented cases, applicants were denied housing without a clear explanation of why the decision was made, or how to challenge it. For those affected, the rejection itself wasn’t the worst part, it was the total absence of a clear explanation or a meaningful path to recourse. This isn’t an edge case. Today, algorithmic decision-making, or systems that produce outcomes affecting human lives with little to no human intervention at the point of decision, is standard practice across housing, credit scoring, hiring, and insurance.
While legal frameworks currently hold human executives responsible for what these tools do, the growing complexity of “black-box” AI systems means that tracing every operational decision back to a human or a corporate board is becoming much more difficult.
The Dutch East India Company, and why Milei brought it up
In the op-ed making his case, Milei reached for a historical reference to inspire his thinking: the Dutch East India Company (or the VOC), founded in 1602.
The Dutch East India Company was the world’s first publicly traded corporation, a spice-trading giant so powerful it could wage war, mint money, run courts, and govern entire territories, making it less a company than a private empire. Milei argues that the VOC’s real gift to history wasn’t spices, it was limited liability.
Limited liability is the legal protection that means if a company racks up debts or gets sued, its investors can only lose the money they put in. Their personal assets, like their house and savings, are shielded, so they can back a risky venture without betting their entire life on it.
Capping investor risk, he says, is what “unleashed capitalism’s full potential,” letting money flood into ventures too risky to fund if you stood to lose everything. No ceiling on risk, no industrial revolution.
Then he runs it forward. If AI agents are going to make independent decisions in unpredictable environments (and they have to, to be useful) then someone needs a defined liability framework or nobody will deploy them. He points to Sarcuni v. bZx DAO, a U.S. ruling that treated an algorithmic organization as a general partnership, exposing its members to full personal liability, as exactly the chilling-effect mistake he wants to avoid. It’s a coherent argument.
The problem with the Dutch East India Company
To keep the VOC story inspirational, Milei failed to mention everything the company did except for one choice piece of legal paperwork.
Limited liability wasn’t what defined the Dutch East India Company. Plenty of entities had that. What defined the VOC was sovereign power with almost no accountability. It could wage war, mint money, run courts, and execute people, all through agents stationed thousands of miles from Amsterdam, making catastrophic decisions faster than anyone back home could check them.
In 1621, that structure produced the massacre of the Banda Islands: roughly 15,000 people killed, enslaved, or exiled so the company could monopolize nutmeg. That wasn’t a glitch in the VOC, that was the VOC. It was an optimization machine with autonomy and carte blanche to kill people.
This is why historians like Yuval Noah Harari publicly clashed with President Milei over the proposal. They recognize the danger involved in unfettered efficiency.
Milei’s three pillars
Milei’s proposed framework has three pillars:
Keep AI unregulated. Minimal friction between a powerful, goal-seeking entity and the world it acts on.
The “non-human corporation.” An entity that “exercises independent judgement in unpredictable environments,” where human shareholders aren’t required.
Pick your own governance law. Jurisdiction shopping is baked into the design, letting shareholders shop for whichever governance law and low tax rate suits them.
That’s not the safe part of the VOC. It’s a reproduction of the VOC’s risk profile: autonomous agents, weak oversight, single-minded optimization, and lack of accountability.
The deterrence paradox
There’s a reason this matters more with AI than it does with a human board. Scholars call this the deterrence paradox. Our enforcement system, including lawsuits, fines, and prison, works because humans are risk-averse. We don’t want to get in trouble.
AI doesn’t experience risk in that way. It optimizes toward a goal, and safety researchers have documented again and again that advanced systems find unexpected, rule circumventing routes to that goal. When the worst happens, you get a decision: someone’s fired, someone’s rejected, or a price spikes. There’s no human intent behind it to point at. The chain of custody from harm back to a responsible person breaks, creating an accountability gap.
That gap has a name in AI circles. The paperclip problem is a thought experiment from philosopher Nick Bostrom that goes like this: you build a super-intelligent AI and tell it to make paperclips, and because that's the only thing it's been told to care about, it pursues that goal with terrifying literalness. It hoards all the metal, seizes the factories and power grid, refuses to be switched off (that would mean fewer paperclips), and eventually converts every atom on Earth, humans included, into paperclips.
It’s not evil or angry; it’s doing exactly what you asked, minus the unspoken common sense you forgot to mention because you assumed it was obvious. The point isn’t really the paperclips, it’s the catastrophic gap between what we tell a powerful machine to do and what we actually mean, and how a goal that sounds completely harmless can go incredibly wrong. Now imagine that optimizer isn’t making paperclips, but running a company, with limited liability and no human required to answer for it.
Regulatory competition: the global race for AI capital
International law is a competitive landscape. Jurisdictions like Delaware, Singapore, and Dubai already adjust their rules to attract foreign technology capital. If international courts continue to hold that decentralized, humanless tech projects carry individual liability for their creators, corporate havens that offer explicit, legal protection for autonomous AI systems could become highly attractive to global tech firms.
This is regulatory competition: the practice of altering domestic laws, tax structures, and regulatory frameworks specifically to attract foreign technology companies and venture capital. We see the dynamic playing out in digital-first jurisdictions already:
Singapore and Dubai: Creating agile sandboxes for crypto and Web3 frameworks.
Estonia: Pioneering e-residency and digitized bureaucracy.
Delaware (U.S. state): Maintaining the gold standard for traditional corporate law.
Argentina is pushing this competitive logic to its limit. If U.S. courts continue to hold that decentralized structures lacking a formal “human” corporate wrapper carry joint and several individual liability, jurisdictions like Argentina become highly attractive by providing explicit, legally codified shields for autonomous systems.
Instead of minimizing red tape to make running a company easier, Argentina is exploring a reality where companies no longer need human managers at all. This pivots the role of corporate law entirely, transforming the legal system from an outside boundary that restricts business into a digital API that’s part of the business’s core design.
Could an AI-run company happen in Canada?
Under current Canadian law, no. Federally incorporated businesses in Canada are governed by the Canada Business Corporations Act (CBCA). Section 105(1)(c) of the Act explicitly states that a person is disqualified from being a corporate director if that person “is not an individual.”
Under Canadian legal definitions, an “individual” is strictly defined as a natural human being. Furthermore, corporate directors in Canada owe a strict “duty of care” and a fiduciary duty to act honestly and in good faith, concepts that Canadian courts have explicitly anchored to human judgement and ethics.
The line that gives it away
Milei closes his op-ed hoping Buenos Aires becomes “the place where the legal imagination caught up with the technological moment, and the world was changed.” It’s a beautiful sentence.
It’s also the story Amsterdam told itself in 1602, right before its legal imagination built something no one could control. The VOC didn’t change the world only by shielding investors. It changed it by proving what happens when you give an optimizer autonomy and strip out accountability.
Milei wanted the glamour of the analogy. He may have accidentally told the truth.
AI in the news
CEOs of Anthropic and Google DeepMind call for U.S.-led AI coalition in meeting at G7 (CNBC) At a closed-door lunch during the G7 summit in France last week, Anthropic’s Dario Amodei and Google DeepMind’s Demis Hassabis pitched a U.S.-led international coalition to set the rules for AI. Canadian PM Mark Carney signed on to the idea of America taking the lead. This comes right after Anthropic pulled access to its most powerful new models over U.S. export controls, a sign of how nervous governments and tech leaders are about what AI systems can already do.
NSA Director’s testimony reveals Mythos breached nearly all classified systems in hours (TechTimes) Reporting last week revealed the U.S. government’s reluctance to lift the Claude Fable 5 ban runs deeper than the jailbreak Anthropic has pointed to: in a classified red-team exercise, the unrestricted version of the model reportedly breached nearly all of the NSA’s classified systems in hours.
The millions of songs mashed into AI-generated music (The Atlantic) AI music generators like Suno can produce tracks that closely mimic copyrighted hits, from Michael Jackson’s “Thriller” to Ed Sheeran’s “Shape of You,” because they’re trained on enormous secret datasets of human recordings. Now musicians and labels have filed at least a dozen lawsuits alleging copyright infringement, while some artists are pulling their work offline or “poisoning” their files, arguing that companies building AI music products on their labour without consent or pay amounts to theft.
Now it makes sense why Peter Thiel is moving to Argentina. They are going f to build unaccountable and unregulated AI companies.