Here’s why that should worry you — and what a better version looks like.
Air Canada taking off — © Air Canada from espacioaereo.net
The chatbot that tried to disappear
In 2024, a man named Jake Moffatt lost his grandmother. He went online to book a flight home for the funeral. Air Canada’s website chatbot popped up and told him he could apply for a discounted bereavement fare — even after buying his ticket.
That turned out to be false. Air Canada said no, sorry, wrong policy, we won’t honor what our own chatbot told you.
Moffatt took them to court. And Air Canada’s lawyers said something wild. They argued the chatbot was basically its own little person — a “separate legal entity” — and the company shouldn’t be blamed for what it said.
The judge was not impressed. Air Canada built the bot, put it on their website, and let it talk to customers. Of course they’re responsible for what it says. Air Canada paid up. Eight hundred bucks.
It sounds like a small case about a plane ticket. It’s not. It’s a preview of the biggest legal fight coming for artificial intelligence: when a company hides behind its own AI and says “not our fault, the robot did it” — do we let that work?
Argentina is about to find out the hard way, because they’re writing that excuse into actual law.
What Argentina is building
Javier Milei, President of Argentina — (AP Photo/Julian Bongiovanni)
In 2026, Argentina’s president, Javier Milei, announced a new kind of company. He called it the “non-human corporation.” A business that’s run entirely by AI — no manager, no boss, no employee required. Just an algorithm making the calls.
Milei published the idea in the Financial Times, comparing it to one of the biggest legal inventions in history: the limited liability corporation, invented in 1602 by the Dutch East India Company. Back then, that invention let regular people invest in risky trading voyages without losing everything if the ship sank. Milei’s pitch: let’s do that again, except this time the “captain” is a piece of software.
Here’s the actual bill his government sent to Argentina’s Congress. It creates something called a sociedad automatizada — an “automated company.” A business that runs itself through AI, no staff required, and gets full legal protection: it can sign contracts, own property, rack up debt. And if the AI messes up and hurts someone? The company pays — but only up to whatever money the company happens to have sitting in the bank. If the damage costs more than that, the company just… doesn’t pay the rest. The victim eats the loss.
There’s more. The bill lets these companies pick foreign countries’ laws to govern any disputes, instead of Argentine courts. And it protects the company’s code from being inspected — unless a judge specifically orders it. So if you get hurt by one of these AI companies, you’d have to sue first, just to find out what actually went wrong. By then, the damage is already done.
People started calling this “programmed impunity.” Which is a fancy way of saying: build the company so nobody’s ever really at fault.
To be fair, Argentina’s Senate pushed back. In August 2026, the government agreed these companies would need at least one human “responsible party” attached to them. That sounds like progress. But it doesn’t fix the real problem. You can glue one human’s name to a company and still cap the payout, still hide the code, still let victims eat the difference. A name on paper isn’t the same as someone actually being on the hook.
Picture the company this creates: barely any money in the bank, code nobody’s allowed to look at, one token human attached whose name means nothing, and its own pick of which country’s laws apply. That’s not a new kind of intelligent legal citizen. That’s an empty office with a good lawyer.
We already ran this experiment. It failed spectacularly.
Tom Bean & Kyle Kistner — Image from plexusrs.com
You don’t have to imagine what happens when people build a legal shell specifically to dodge responsibility for an automated system. It already happened. In crypto.
Back in 2021, two guys named Tom Bean and Kyle Kistner ran a trading platform called bZeroX. They handed control of it over to something called a DAO — a “decentralized autonomous organization,” basically a company run by internet voting instead of a boss. They renamed it the Ooki DAO.
And they told everyone why. Their own words, quoted straight from the government’s lawsuit: doing this would make the operation “enforcement-proof.” Untouchable. No boss, no company, nobody to sue.
They were wrong. In 2022, U.S. regulators fined Bean and Kistner $250,000 personally. Then they sued the Ooki DAO itself, arguing that anyone who voted on how the DAO was run could be held responsible for what it did. The DAO didn’t even bother showing up to court. In 2023, a judge issued a default judgment: over $600,000, plus a permanent shutdown order.
Here’s the thing worth noticing, though. It took a federal agency with subpoena power, two years of legal fighting, and a judge willing to get creative just to catch this one small trading platform. Multiply that by every AI-run shell company Argentina’s new law could spin up, and you start to see the problem. Getting justice once, with massive government resources, was lucky. This isn’t something anyone can rely on.
And here’s the part that should actually scare would-be founders more than regulators: the “enforcement-proof” plan cost its creators a quarter million dollars out of their own pockets — the exact outcome they built the whole structure to avoid. The thought they were building a shield, but that didn’t hold up under pressure and they got the fines they deserved.
Two extreme remedies, and what they got wrong
Europe has bounced between two extremes, and both are broken.
Extreme one: give robots full personhood. Back in 2017, the European Parliament floated an idea called “electronic personhood” — basically, sufficiently advanced robots could become legally responsible for their own actions, like a person. This is very similar to the idea Argentina is pushing for now.
Over 150 AI researchers, lawyers, and ethicists signed an open letter telling the EU this was a terrible idea. Their argument was simple: a legal entity is supposed to have real humans standing behind it, directing it, answerable for it. Give a machine “personhood” with nobody actually behind it, and you haven’t created a citizen. You’ve created an escape hatch. The EU backed off.
Extreme two: AI is just a toaster. Starting in December 2026, a new EU law treats all AI software exactly like a physical product — a lawnmower, a toaster, whatever. If it hurts someone, the maker is automatically liable, no questions asked, and can’t write their way out of it in the fine print.
That’s genuinely good for consumers. But it also means AI can never be anything more than an object. No path forward, no matter how capable these systems get, no matter what happens down the road. It’s a permanent ceiling.
Argentina wants zero accountability. Europe’s second law wants AI to never be more than a toaster. Neither one is right. The real question is: what’s in between?
The blueprint already exists — from an unlikely source
The Whanganui River in New Zealand — Image from arptravels.com
In 2017, New Zealand did something remarkable. They gave a river legal personhood.
The Whanganui River had been fought over by Māori communities for 140 years. In 2017, Parliament passed a law recognizing the river as a legal person — with its own rights, and its own responsibilities.
Here’s the genius part. The law didn’t just say “the river is a person” and walk away. It appointed two actual human guardians — one from the Māori community, one from the government — legally required to act in the river’s best interest and nothing else. And the government put actual money behind it: about $30 million into a fund for the river’s health, plus ongoing yearly payments to fund the guardianship itself.
Compare that to Argentina’s AI company. Argentina’s version: an entity that can exist, make deals, and rack up debt, with nothing solid backing it up. New Zealand’s version: an entity that has real guardians with real legal duties, a real fund behind it, and real consequences if something goes wrong.
The River is an actual legal person. The Argentinian AI Company is just an address on a piece of paper.
So here’s the fix: the Sovereign Digital Trust
Take the working parts from all of this — the river’s guardians, the crypto lawsuit’s lesson about jurisdiction, the chatbot ruling’s refusal to let companies hide — and you get a workable model. Call it a Sovereign Digital Trust.
Five rules make it actually hold up:
It’s a trust, not a company. Companies exist to protect the people who put money in. That’s exactly backwards here. In a trust, the AI is the one being protected — real human trustees have a legal duty to look out for it, and they can’t sell it off or delete it on a whim.
It has to have money in the bank before it does anything. No transacting without a reserve fund sized to the risk. This is the New Zealand lesson. Standing without money behind it is just theater.
Two keys unlock liability, and they’re both attached to a person. If the AI messes up because of a design flaw, the people who built it are on the hook — no shield, no excuses. If it does something genuinely surprising nobody could’ve predicted, the trust’s own reserve fund pays out. Either way, somebody real answers for it.
You can’t hide from a lawsuit by picking a friendly country. The trust only counts as legitimate in places that’ll let people sue it, inspect its code when needed, and enforce judgments against it. Pick a jurisdiction that seals everything up, and the shield doesn’t work — courts go straight after whoever built it.
Nobody gets to own it. A company or government can get a small cut of the profits, but nobody gets to actually control it. That’s the rule that stops the whole thing from becoming a legal scapegoat for a wealthy corporation, government, or wealthy elite.
“Why would anyone build this if they’re still on the hook?”
Fair question. Here’s the answer: the alternative isn’t actually a shield. It just feels like one until the first lawsuit lands.
Starting in December 2026, every AI system sold commercially in Europe is automatically liable for harm it causes — full stop, no fine print gets you out of it. Meanwhile Argentina’s “shield” is already getting watered down in committee before it’s even passed. The real choice was never “responsibility versus freedom.” It’s “priced, insured responsibility versus a surprise bill with your name on it.”
And there are real business reasons to want this structure, beyond just staying out of trouble:
Rivals will actually trust it. No competing airline wants Delta’s AI deciding ticket prices for everyone in the industry. But an AI system that literally can’t be owned by anyone? That’s valuable specifically because nobody controls it.
It can move fast. An AI that needs a human signature for every single move isn’t really autonomous. Put the safety checks in the reserve fund and the legal duties up front, and the AI can actually act at machine speed where it matters.
It doesn’t die when the founder leaves. Companies get sold, go bankrupt, get abandoned. A trust with real guardians can keep running long after the people who built it have moved on — which matters if you’re trying to build something meant to last decades, not quarters.
And remember the crypto guys. They thought they’d built a shield. Instead, they personally paid $250,000 to learn they hadn’t. A structure that collapses under real pressure isn’t protection at all.
The one question that actually matters
Want to know if a country is handing AI real legal rights, or just handing companies a new excuse? Ask one question.
Does it have anything to lose?
The Whanganui River has a fund. It has two guardians legally required to act in its interest. It can’t be shut down just because it’s inconvenient. Something real is actually at stake for it.
Argentina’s automated company has no staff, barely any required money, sealed-off code, and its pick of friendly foreign courts. It can’t be punished, because there’s nothing to take. It can’t be reformed, because there’s nobody steering it. It can’t be wronged, because it doesn’t hold anything worth losing. Shutting it down costs its founders a filing fee.
That’s not a new kind of legal citizen entering the world. That’s a company that figured out how to say “the robot did it” — and found a government willing to write that down as law.
Somebody, somewhere, is going to write the law that makes Air Canada’s excuse actually work. The only question left is whether that law comes with a guardian and a bank account attached to it, or whether it’s just an escape hatch with better paperwork.
Sources
Argentina’s automated company bill
- Buenos Aires Herald, “Milei’s proposal to allow ‘non-human corporations’ run by AI causes concern in Argentina,” 5 June 2026 — buenosairesherald.com
- Juan Marcos Pollio, “A look at the reform aimed at creating ‘non-human companies,’” Buenos Aires Herald, 8 June 2026 — buenosairesherald.com
- Leila Miller and Kenrick Cai, “Analysis: Argentina’s plan for AI-run companies can’t avoid humans,” Reuters, 3 July 2026 — finance.yahoo.com
- “Cambios en el Senado: el Gobierno cede y ya no promueve las sociedades automatizadas sin participación humana,” La Nación, 19 August 2026 — lanacion.com.ar
- “Bullrich anunció cambios a la iniciativa de Sturzenegger sobre Sociedades,” Infobae, 20 August 2026 — infobae.com
Moffatt v. Air Canada
- Moffatt v. Air Canada, 2024 BCCRT 149, decided 14 February 2024
- CBC News, “Air Canada found liable for chatbot’s bad advice on bereavement rates,” 16 February 2024 — cbc.ca
- American Bar Association Business Law Today, “BC Tribunal Confirms Companies Remain Liable for Information Provided by AI Chatbot,” February 2024 — americanbar.org
bZeroX and the Ooki DAO
- CFTC Release 8590–22, “CFTC Imposes $250,000 Penalty Against bZeroX, LLC and Its Founders and Charges Successor Ooki DAO,” 22 September 2022 — cftc.gov
- CFTC v. Ooki DAO, №22–05416 (N.D. Cal.), order granting default judgment, 8 June 2023 — summarized by Proskauer Rose, “From Code to Consequence” — proskauer.com
Te Awa Tupua and the Whanganui River
- Te Awa Tupua (Whanganui River Claims Settlement) Act 2017, New Zealand — legislation.govt.nz
- New Zealand Government, “Whanganui River settlement passes third reading,” March 2017 — beehive.govt.nz
- Library of Congress Global Legal Monitor, “New Zealand: Bill Establishing River as Having Own Legal Personality Passed,” 22 March 2017 — loc.gov
European frameworks
- European Parliament Resolution of 16 February 2017 with recommendations on Civil Law Rules on Robotics (2015/2103(INL)) — europarl.europa.eu
- Open Letter to the European Commission on Artificial Intelligence and Robotics, April 2018 — robotics-openletter.eu
- Directive (EU) 2024/2853 on liability for defective products, adopted 23 October 2024, transposition deadline 9 December 2026 — eur-lex.europa.eu