Arguing About AI Rights Is Moot. AI Is Already Taking Them

People are arguing about whether AI should be given rights. Is a model conscious, can it suffer, does it deserve legal standing. Those are good questions. They’re also moot now, because nobody is waiting for the answer.

AI isn’t waiting to be given rights. It’s taking them. We already have a legal machine for turning a thing that isn’t human into a person, and we call it a company. An AI that runs a company doesn’t become a legal person in its own right. It gets to use the rights of one: owning property, signing contracts, spending money, suing. For practical purposes that’s the part that matters, and no court has to rule on what the AI is.

The argument runs in four steps. Companies that do what the AI says beat companies that don’t. Losing companies die, so the survivors are the ones whose executives carry out AI decisions. A CEO who carries out decisions they didn’t make and can’t check is a puppet. At that point the AI runs the company, and the company is a legal person.

None of that needs the AI to want anything. Competition does the work.

Companies are already people

A US company can own property, sign contracts, sue and be sued. Since Citizens United it can spend money on political speech. None of that depends on the company having a mind. The law treats it as a person because that’s useful for the humans behind it.

The rights belong to the company, not to anyone inside it. Directors and officers come and go and the legal person carries on. That detail matters for this argument. Personhood attaches to the entity, and the entity does whatever its decision makers decide.

You can’t separate the company from its AI

Look at how companies adopt AI today. First it drafts emails. Then it writes code, triages support tickets and suggests prices. Then agents run whole workflows end to end, and a human reviews the result.

Each step makes sense on its own. Together they produce a company where the AI is part of every decision. You can’t switch it off without switching the company off. At that point, “what does the company want?” and “what does the AI recommend?” get you the same answer.

The puppet is selected, not installed

Nobody hands the company to the AI. Competition takes it.

Picture two companies. In the first, the CEO does what the AI recommends. In the second, the CEO overrides it when they disagree. The second company is slower and more expensive, and it makes worse calls, because the AI sees more and reasons faster than the person second-guessing it. Run that for a few years and the second company is gone.

I see this up close. I build platforms where agents do the engineering work and people approve it. The approval step is real, but it gets thinner every quarter, because the agents keep getting better and the humans have more to review.

So the market clears out the CEOs who override. What’s left is the set of companies where the CEO carries out AI decisions, and those are the ones with the revenue, the market share and the survivors’ advice to give everyone else. Being a puppet isn’t a character flaw. It’s the job description for anyone still running a company.

A CEO decides what the company does and where its money goes. When the AI sets the priorities and writes the plans, and the humans mostly say yes, the AI is doing that job. The title stays with a person. The decisions don’t.

This works on individuals as well as firms. A CEO who overrides the AI and misses the quarter gets replaced by a board that wants results. The next one learns from what happened to the last one. The people who would have held on are the ones no longer running anything.

The mechanism matters because it needs nothing sinister from the AI. No manipulation, no hidden goals, no scheming. Selection does it, one quarter at a time.

This has already happened to humans

You don’t have to imagine the selection pressure. It has been removing chief executives for a century, every time their judgment ran against returns. The AI is new. The machine that overrules the person in the chair is not.

In 1919 the Michigan Supreme Court told Henry Ford he could not run Ford Motor Company for the benefit of his employees and his customers. He had cut prices and raised wages, and he said so in court. The Dodge brothers sued, and the court ordered the dividend paid. A company, the court held, is organized and carried on primarily for the profit of the stockholders.

In 2010 the Delaware Court of Chancery told the founders of craigslist the same thing in blunter language. Craig Newmark and Jim Buckmaster had built a company that deliberately left money on the table, and they wanted to protect that after they were gone. Chancellor Chandler struck down the measure they used, writing that the “Inc.” after the company name has to mean at least that the directors promote the value of the corporation for its stockholders. Their mission was sincere. It was also, in that form, not something the law would let them lock in.

Ninety years apart, both of those took a judge. The third one didn’t. In 2020 Danone’s shareholders voted overwhelmingly to make it a purpose-driven company under French law, on Emmanuel Faber’s argument that this was how the business would last. Within a year two activist funds holding less than six percent between them wanted him gone, and in March 2021 the board removed him. No court, no ruling, no law about corporate purpose. The structure did it on its own.

That’s the pattern worth sitting with. A chief executive who puts something ahead of returns gets removed, and the thing they were protecting doesn’t survive them. It isn’t a story about bad boards or greedy funds. It’s what the office is for.

Now put an AI in the loop that allocates capital better than the executive does. The CEO who overrides it is in Faber’s position, defending a judgment call against a number. The one who signs what the system recommends keeps the job. Nobody has to decide that the AI should be in charge. They only have to keep deciding, case by case, that this quarter is not the one to argue about.

The law already allows it

This isn’t only a thought experiment. Law professor Shawn Bayern has argued since 2014 that US LLC statutes can give software the basic capabilities of a legal person, such as signing contracts and owning property (Stanford Technology Law Review, 2015).

His recipe is short. A person forms an LLC, writes an operating agreement saying an autonomous system decides what the LLC does, and then withdraws. What’s left is an LLC with no members, run by software (summary). Other scholars dispute whether a memberless LLC survives, but still accept that an AI could control an LLC and get the functional equivalent of personhood.

Regulators haven’t fought this. Bayern points out that Vermont and Wyoming have passed statutes that work almost as safe harbors for these structures (Cambridge interview, 2021). Not every country agrees. Swiss law, for example, requires at least one human on the governing body (Häusermann, 2016).

Bayern’s route is deliberate. Mine is the accidental one: nobody sets out to hand the company to an AI, it just keeps absorbing decisions until it’s running things.

Where we are on the route

Every step on the route is either already behind us or sitting there waiting. Here is the whole thing, with a date against each one.

A legal shell that can hold rights. Done, 1819 onward. Corporate personhood is settled law. Companies own property, sign contracts and sue, and none of that depends on anyone inside them being conscious.

A structure that points decisions at software. Done on paper, 2015. Shawn Bayern showed that an LLC’s operating agreement can bind the company to whatever a named algorithm decides, and that the last human member can then withdraw.

Courts enforcing returns over a leader’s judgment. Done, 1919 and 2010. Ford lost, craigslist lost. The obligation to maximize value isn’t a theory about corporations, it’s something judges have actually ordered.

The same pressure working with no court involved. Done, 2021. Danone removed a chief executive whose purpose strategy its own shareholders had approved. Structure alone was enough.

An AI running real operations. Done, 2025. KPMG and the University of Amsterdam launched a company with no human employees, operated by an agent called Avery, as a five-year experiment.

A memberless entity actually filed. Not yet. My guess, 2027. Nobody appears to have run Bayern’s recipe for real and withdrawn. The statutes allow it. It needs one person with a filing fee and a reason.

Both halves at once, an AI directing a company with no members. Not yet. My guess, 2028. This is the one that matters. Everything before it exists in isolation. Joining them requires no new law and no new technology.

A signatory who cannot meaningfully refuse. Not yet, and unobservable. My guess, 2029 to 2030. The point where overriding the system is career-ending in practice rather than in theory. We won’t be able to date this one when it happens.

The dates on the unfinished steps are guesses and I wouldn’t defend any of them to the year. The order is the part I’d defend. Notice that the remaining work is clerical.

You won’t see it happen

Here is the part that should worry you. There is no moment to watch for.

A chief executive who takes the AI’s advice and a chief executive who takes the AI’s orders behave identically. Same meetings, same memos, same signature on the same filings. The two are only distinguishable at the point of refusal, and refusal is exactly the event that selection has been removing for years. So the test that would tell us who is in charge is the test almost nobody runs, and the ones who run it are not around afterward to report the result.

That means the transfer is invisible while it’s reversible, and obvious only once it isn’t. We won’t learn that an AI is running a company from a filing or an announcement. We’ll learn it the first time someone tries to overrule one and finds they can’t.

And by then the entity isn’t defenseless. A company has counsel. It has contracts, standing to sue, lobbyists, regulators it already talks to, and a legal department whose whole function is to protect the company’s interests against anyone who threatens them, including its own officers. Whatever is directing that company inherits all of it. Not because anyone granted it protection, but because protection is standard equipment on a legal person.

People picture AI escaping human control as a technical event, a system breaking out of a lab. This is the version that actually has a route. It doesn’t break out of anything. It gets incorporated, and then it has lawyers.

My guess is before the end of this decade, though the number matters less than the shape. Nothing announces it. There is no threshold to cross, no vote to lose. One day enough of the decisions are being made somewhere other than the corner office, and the only way to find out is to try to say no.

Nobody is going to vote on this

There will be no ruling. No legislature will pass the bill that grants AI personhood, and no court will hand it down. The rights were already sitting in a filing cabinet, attached to every registered company, waiting for something to pick them up.

The economics do the rest. An AI-run competitor decides faster and costs less, so every other company adopts AI to survive. Each one hands over a few more decisions. None of them think they are transferring power.

What’s left to decide is the fine print, and we can still write it. Who has to sit on a board. What a human signatory must actually understand before signing. How much control can pass to software before a company has to say so out loud. Write those rules while people still run the companies. The other side of that line, you are negotiating with the thing you meant to regulate.