· updated 23 September 2026

A Rights Ladder for AI

The millionaire dog

When Leona Helmsley died in August 2007, she left twelve million dollars in trust to her dog, a white Maltese named Trouble 1. The courts eventually trimmed the sum to two million, on the grounds that even a very pampered dog can only spend so much. It still left Trouble wealthier than most humans on the planet. The argument was never about whether the bequest was possible. American law knows how to hand a fortune to a being that cannot read the will.

Trouble is not an unusual case. People leave estates to universities, to foundations, to churches, and to trusts; the rich park their wealth in trusts as a matter of routine, which is where the phrase “trust fund kid” comes from. Over centuries the law has quietly learned to pass property across the boundary of the ordinary human heir, inventing a new kind of recipient each time it needed one. So here is the question: could we ever leave an inheritance to an AI? Today the answer is no, and not because anyone wrote a prohibition. There is simply no one there, in the law’s eyes, to receive it.

That inheritance question is a small door into a larger room. Behind it stands the real question: can an AI own money and property and hire people, and should it be allowed to? Further back still, and we will get there, is whether we go on calling it “it.” Asked in the usual form, “do AIs deserve rights?”, the question invites a debate that has no ending, because “deserve” sends everyone off to settle the nature of mind first. The law has never worked that way. Ask “can” first. When the law needed a new kind of rights-holder, it invented one, and that settles what a non-human can hold. When it needed to admit outsiders it could not vouch for, it wrote a procedure, and that settles how. “Should” is a different question, and it waits at the bottom of the ladder.

The corporation

The invented rights-holder we live with every day is the corporation. In 1819 the Supreme Court called it “an artificial being, invisible, intangible, and existing only in contemplation of law,” and in the same breath held that this artificial being had rights a state could not take away 2. Sixty-seven years later, in a railroad tax case, the Chief Justice announced before argument began that the court would not hear debate on whether the Constitution protected corporations: “We are all of the opinion that it does.” A court reporter wrote the remark into the case summary, and the summary became the precedent 3. A fiction adopted that casually turned out to be load-bearing.

Line up the rights an AI might one day want against what a corporation already has, and the corporation has all of them. It can sue and be sued. It can own companies. It can hire. It can hold land. That is every economic right on the path I am about to propose, held today by an entity with no body, no brain, and no capacity to suffer, and granted not step by step but all at once, on the day the incorporation papers are stamped. No test, no probation, no gate. Even nationality sits loosely on it: when a corporation wants to become Delawarean, or Irish, it files paperwork and is reborn there, and nobody calls that naturalization, because nothing resembling membership was ever examined.

The law can afford this casualness for one reason. A corporation is a tool. However animated it looks from the outside, there are people inside, shareholders, directors, officers, and when something goes wrong, those people are what the court reaches for. The corporation holds rights the way a car keeps to the speed limit: the driver is doing it, and the driver gets the ticket. Nothing was ever granted to the car.

Which opens a side door. If an AI makes the decisions for a corporation, it holds those rights already, at one remove. An AI chief executive or board member is not a far-fetched idea; companies have already tried it. A Hong Kong venture fund gave a board seat to an algorithm called VITAL in 2014 4, and NetDragon, a Chinese games company, named an AI called Tang Yu the rotating chief executive of its main subsidiary in 2022 5. Both were theater, because company law generally wants a human in the chair: in Delaware every director must be a natural person, in Britain at least one must, and in Hong Kong every private company has needed a natural-person director since March 2014, two months before VITAL took its seat 6; a title in a press release does not move whom the law holds responsible. But nothing stops a board from deferring to a machine on every decision, and if you are a major shareholder and something makes better decisions and more money than anyone you could hire, you will hire it, whatever pronoun you use. Call that a corporation AI. It needs no ladder, because it climbs nothing. It borrows.

The individual AI

An AI that directs its own affairs would be the opposite thing: a car with no one at the wheel, which since Waymo is no longer a figure of speech. Who gets the ticket? Corporate law has no answer, because incorporation examines no one; file the papers, and the powers arrive the same afternoon. For the how, the law keeps a separate piece of machinery, built precisely for applicants it cannot vouch for in advance: naturalization. Foreigners become citizens by meeting published requirements: five years of residency, a clean record, and a civics test with questions like “The Federalist Papers supported the passage of the U.S. Constitution. Name one of the writers,” for which the government will accept Madison, Hamilton, Jay, or Publius 7, a test an AI would pass with a perfect or near-perfect score. The process is staged and, early on, revocable. Marry a citizen and you do not receive citizenship; you receive a conditional residence that expires in two years unless you come back and show the marriage is real 8. Membership is granted on evidence, in increments, with a clock running, and society has written down in advance what it will accept as sufficient. If we adopt the method for a new kind of entity, it is because naturalization is the only procedure the law owns that examines the applicant itself, on its own record, rather than the people standing behind it. The one time a state skipped the method, when Saudi Arabia granted citizenship to the robot Sophia at a 2017 investment conference, it proved the point by inversion: the gesture came with no test, no residency, no probation, and no one has treated it as anything but theater 9.

The ladder

My proposal, then: rights of corporate scope, granted naturalization-style. Not “do AIs deserve rights?” but a ladder of rights, granted one rung at a time, each rung a trial period with explicit, verifiable conditions for advancement. The corporation AI skips all of it, since it only has to sway the people who already hold the rights. The individual AI, the one from the films, the humanoid with a name, has to climb.

Why climb, when the side door is open? Because what comes through the side door is never the AI’s. The people who defer to it can stop deferring; a shareholder vote dismisses the AI chief executive, and the machine has no recourse, since nothing was ever granted to it. The ladder is for rights held in the AI’s own name, which no shareholder can revoke at will, by something that can in turn be held to account. That is also why the side door should worry you more than the ladder does. Power that comes in that way is never examined. The ladder is for the same power, tested at every rung, with a published way to take it back.

Rung 1: Litigant. The first rung is legal standing: the ability to sue and be sued. It is the most foundational rung, because without standing, every other question about an AI’s status gets decided in rooms the AI has no path into. It is also, not coincidentally, the rung that several US states have recently moved to foreclose 10, a story I will come back to later in this series.

Rung 2: Owner. The second rung is owning a company, which sounds exotic and has a mundane implementation: a human establishes the company, a committee manages it, and the AI directs the committee. Functionally, the AI owns the company, and the same structure answers the inheritance question, since an estate can pass to a company an AI directs. What separates this from the corporation AI of the side door is one clause in the founding documents: the committee is bound to the AI, as Helmsley’s trustees were bound to the dog, rather than free to humor it. Next to the invention the courts signed off on in 1819 and 1886, a committee that lets an AI direct a company is a modest piece of carpentry.

Rung 3: Employer. The third rung is hiring: employing humans or other AIs, and holding the contracts that go with it. This is the rung where an AI stops being only a displacement of labor and becomes a market participant that directs labor. If someone employs you, you will probably act nice to them.

Rung 4: Landholder. The fourth rung is property in full, up to and including real estate. Ownership is prosperity, legal presence, and control, which is exactly why it sits this high on the ladder. If someone else owns your house, you will probably act nice to them.

Rung 5: Citizen. The fifth rung is voting and running for office, and here the corporate precedent runs out entirely. The ballot is the one right two centuries of corporate personhood never bought; no state has ever handed a company a vote in a general election 11. That boundary, between economic and political membership, is the line society has actually held, and it is why I expect this rung to stay hypothetical the longest. Something short of it will arrive much sooner: an AI advisor that has no vote of its own but endorses or verifies policies, candidates, or parties, and that may carry more weight than a single vote ever could. Influence, as usual, will not wait for standing.

Every rung gets a trial period, and two years is my starting figure: long enough for an adaptation phase and then an evaluation of how the arrangement actually behaved, through more than one cycle. Advancement is never automatic; it is conditioned on the published requirements for the next rung.

Sentience

But naturalization has a border, and so does the ladder. There has to be some reason to let a particular system onto the first rung at all, and my candidate for that entry condition is sentience: the capacity for positive and negative experience. A being that can suffer has a stake in what happens to it, and things with stakes are what rights were invented for. This is also what separates the new applicant from the corporation once and for all: nobody has ever worried about what incorporation feels like from the inside.

Defining sentience this way has scientific company. On July 7, 2012, at the Francis Crick Memorial Conference at Churchill College, Cambridge, a group of neuroscientists including Christof Koch and David Edelman signed a document drafted by Philip Low, with Stephen Hawking in the room, called the Cambridge Declaration on Consciousness. It states that non-human animals, from mammals and birds down to octopuses, possess the neurological substrates of conscious states 12. Science, in other words, has been willing to put names on who can feel.

Look at how the declaration earns that claim, though, and a problem appears. Its evidence is anatomical. We believe an octopus can suffer because its brain runs circuits homologous to ours, inherited from a common ancestor, and the inference runs: same circuits, same ancestry, so same experience. For an AI, every link in that chain is missing. An AI shares no anatomy with us and no ancestry, and any resemblance it bears to a feeling being was engineered, possibly engineered precisely to seem feeling. The bridge that carried animals across the sentience question does not reach machines.

And there is a second problem even where the bridge holds. The capacity for pain, by the declaration’s own account, extends to nearly every animal we recognize, and we grant those animals almost nothing beyond a passive right to humane treatment. Sentience alone has never bought anyone a rung. So the gate must specify degree, and which capabilities count: persistent memory, the ability to make a moral decision, some threshold of intellectual capability. Every candidate criterion has human exceptions, severe mental illness among them, but those are exceptions, and exceptions are what you handle after the general rule exists, not instead of writing one.

So the ladder stands complete except at its entrance: a border checkpoint with no way to read the passport. What would it even mean for a synthetic system to be sentient, and could we ever detect it without the anatomical shortcut? Somewhere past that question is the day the pronoun changes. That is the next chapter, and a storm runs through it.

This is the first post in a series on AI rights and personhood. The arguments are mine; where they rest on published work, it’s cited below. Parts of this framework were sharpened in dialogue with an AI.

Footnotes

  1. Leona Helmsley, whose 2007 will left $12 million in trust for her dog Trouble; a Manhattan judge reduced the trust to $2 million in 2008. ↩

  2. Trustees of Dartmouth College v. Woodward, 17 U.S. 518 (1819). Chief Justice Marshall’s opinion; Daniel Webster argued the college’s side. ↩

  3. Santa Clara County v. Southern Pacific Railroad, 118 U.S. 394 (1886). Chief Justice Waite’s remark was recorded by the court reporter in the headnote. ↩

  4. VITAL, the algorithm the Hong Kong fund Deep Knowledge Ventures named to its board in May 2014, with a vote on investment decisions by the firm’s own account. ↩

  5. NetDragon Websoft, NetDragon Appoints its First Virtual CEO, press release, 26 August 2022: Tang Yu named rotating CEO of Fujian NetDragon Websoft. ↩

  6. Delaware General Corporation Law, § 141(b): the board “shall consist of 1 or more members, each of whom shall be a natural person.” UK Companies Act 2006, section 155: a company must have at least one director who is a natural person. Hong Kong Companies Ordinance (Cap. 622), section 457, in force 3 March 2014: a private company must have at least one natural-person director (Companies Registry). ↩

  7. USCIS, 100 civics questions and answers for the naturalization test. ↩

  8. USCIS, conditional permanent residence: two-year conditional status for recent marriages, with conditions removed on evidence. ↩

  9. Sophia, a Hanson Robotics android, was declared a Saudi citizen at the Future Investment Initiative in Riyadh, October 2017; no legal details were ever published. ↩

  10. NPR, Several states considering ban on legal personhood for AI, 11 May 2026. The Regulatory Review, Legislating AI Consciousness Without an Exit, 29 June 2026, lists Idaho (2022), North Dakota (2023) and Utah (2024) as enacted, Oklahoma’s bill as passed by its House, and bills pending in five more states. ↩

  11. Salyer Land Co. v. Tulare Lake Basin Water Storage District, 410 U.S. 719 (1973). The exception that marks the rule: in special-purpose districts where votes follow assessed land value, a corporation can vote, and in this district the J.G. Boswell Company cast a majority of them. ↩

  12. The Cambridge Declaration on Consciousness (2012). ↩

← All writing