TL;DR: The standard objection to AI-run companies: a machine cannot go to jail. Neither can a corporation, yet we trust one because four things are bolted on: owners who can lose, assets a court can seize, a revocable charter, and a named person who answers. That human is the hinge: the one whose fear improves the next decision. For AI-run companies: a suable entity with a full log beats an anonymous agent. Against: a person feels risk before the harm, while bankruptcy comes after. Argentina and Delaware both kept someone answerable, and in Delaware, that someone may be a company.
Want to listen to this article?
Orion Playbook newsletter subscribers can listen to the AI-generated audio version for free. Already a subscriber? Log in here
Asked if he would let a better AI run Shopify, its CEO, Tobi Lütke, said of course, then named the obstacle. Nobody would have recourse, because a machine can't go to jail. It is a good line that also describes every corporation you have worked for.
In 1819, the U.S. Chief Justice called the corporation "an artificial being, invisible, intangible, and existing only in contemplation of law," which is a lawyer's way of saying imaginary. We let the imaginary being own property and be sued, and the economy runs on it.
The retort is too easy: the corporation has people inside. Directors can be sued, disqualified, and, when it goes badly enough, imprisoned. The company can't go to jail. Its CEO can. In Germany, a company cannot even commit a crime; its fate hangs on what a manager did. Company law's great trick is that a fictional person takes the blame while a real one, with a home address, loses the sleep.
In 2026, Argentina's Javier Milei offered to retire the real one. In his pitch, "human shareholders may participate, but are not required." Yuval Noah Harari objected that prison means nothing to an AI. Milei turned Harari's own point against him: if bankruptcy would be, for an AI, "the equivalent of death," it has more to lose than any executive. Motivation by fear of extinction was, until now, mostly a technique for quarterly reviews. Both cited the Dutch East India Company: Milei as the birth of limited liability, Harari as a company that became a state. Same ship, two ports.
So what did we bolt onto the corporation before we trusted it? I count four attachments: owners who can lose their stake, assets a court can seize, a charter the state can revoke, and a named person who answers. The first three are levers. The fourth is the hinge, and every grant of legal personhood I can find keeps one. When New Zealand made a river a legal person in 2017, the same statute appointed it a "human face." Even the waterway needed a spokesperson.
Personhood is a grant, which means it can be granted again. Idaho banned personhood for AI in the same sentence as animals and inanimate objects, then exempted the corporation by name.

Two forces have defined the tension between those attachments since the first trading companies. Limited liability moved owners out from under the risk, so strangers would fund a voyage. Directors' duties moved managers back under it, so someone would care how the voyage went. Each boom loosens the rope, each scandal tightens it, and nobody has ever won. A company run by AI is limited liability pulled all the way: the levers stay, the hinge goes.
The hinge matters because it closes a loop. A decision creates a risk, the risk reaches someone who can lose something, and the fear improves the next decision. The working part is a person, awake at 3 a.m. Take that person out and the only alarm left is the balance sheet, which only shows up when it's too late. Every agent you deploy runs this test in miniature: it takes the decision, and the loop must find its person elsewhere.
The case for trying is serious. An AI's legal shell gives a harmed party something to sue, seize, or dissolve, which beats a shrug from a chain of vendors. It can log every decision, a richer record than board minutes, which counsel drafts to reveal little. The human hinge also creaks: directors carry insurance.
The objection is timing. A person feels risk before the harm; bankruptcy arrives after it, and a thinly funded AI company could die cheaply while its founders stay shielded. One law professor asks whether an AI would learn not to get caught, which is also a fair summary of the history of compliance.
A second loop runs outside the company: personhood follows usefulness, and the places that grant it compete for business. Delaware, in the business of selling charters, has floated its own AI-run company. Argentina now promises a flesh-and-blood person in control; Delaware's draft lets its one required owner be "a person or an entity." The human we keep insisting on may be another corporation, which moves the question up a floor without answering it. Insurers have noticed that a directors-and-officers policy for a company with no directors or officers is an unusual product.
None of this is law yet; a human still answers today. The loop only works if that human knows it is them, and nobody volunteers for 3 a.m. Ask each team to name who a regulator would call about the decisions its agents shape, and have your AI ask for that name before it acts.
Our comfort with the corporation was built one scandal and one statute at a time. An AI company could plausibly carry three of the four attachments. Nobody has found a substitute for the fourth yet. So the question for your board, and for whoever drafts the next law: what would have to be true for us to be as comfortable with an AI holding responsibility as we are with a corporation holding it today?

All the images were generated with AI (ChatGPT Images, Gemini Nano Banana, Claude Opus) by Gérard Métrailler.
Food for your AI
Paste this into your AI's standing instructions, so every decision it takes arrives with a name attached.
Before you take any action that commits money, makes a promise to a customer, agrees to terms, or sends data outside this team, tell me the name of the person who answers for that decision. If no one has been named, stop and ask me who it is, and do not proceed until I tell you. At the end of each week, list every decision you took or shaped, with the owner beside each, and put any decision that belongs to no one at the top.
Sources
Gobierno de la República Argentina, Presidencia de la Nación. "Comunicado Oficial Número 149." June 18, 2026. https://www.argentina.gob.ar/noticias/comunicado-oficial-numero-149. Accessed 2026-11-12.
Idaho Code § 5-346, "Personhood Status." Added 2022, ch. 322. Idaho State Legislature. https://legislature.idaho.gov/statutesrules/idstat/Title5/T5CH3/SECT5-346/. Accessed 2026-11-12.
Nay, John, and Charuni Patibanda-Sanchez. "Exclusive: Delaware proposes testing the AIC, a new legal entity for agents in a regulatory sandbox." Fortune, July 14, 2026. https://fortune.com/2026/07/14/exclusive-delaware-ai-agents-legal-entity-proposal-llc-pbc/. Accessed 2026-11-12.
Te Awa Tupua (Whanganui River Claims Settlement) Act 2017, No. 7. New Zealand Legislation. https://legislation.govt.nz/act/public/2017/7/en/latest/. Accessed 2026-11-12.
Trustees of Dartmouth College v. Woodward, 17 U.S. 518 (1819). Legal Information Institute, Cornell Law School. https://www.law.cornell.edu/supremecourt/text/17/518. Accessed 2026-11-12.


