In 2005, an Asian elephant named Happy did something only a handful of species had ever managed: she recognized herself in a mirror. Researchers marked her skin and watched her touch the mark on her own body, in front of the glass, not on the glass — evidence, under the mirror self-recognition test used across species from humans to apes to dolphins, that she could use the reflection to locate a mark on her own body.
Seventeen years later, in 2022, lawyers stood before New York's highest court arguing that this made Happy a legal person — someone entitled to challenge her confinement, alone, in a Bronx Zoo enclosure, through a writ of habeas corpus.
The court's answer is worth reading closely, because of what it doesn't dispute. "No one disputes that elephants are intelligent beings deserving of proper care and compassion," the majority wrote. And then it ruled against her anyway — not because the evidence of her mind was weak, but because habeas corpus "is intended to protect the liberty of human beings," full stop. One judge went further, warning that ruling for Happy "would have an enormous destabilizing impact on modern society."
Read that sentence again. The court did not say Happy failed a test of capacity. It said that allowing her to invoke habeas corpus would be too disruptive to how humans currently use animals. The gate didn't stay shut because she couldn't demonstrate a mind. It stayed shut because too much was already built on the assumption that she didn't have one. This does not mean the court consciously weighed social convenience against evidence of cognition; rather, the existing legal categories made that consequence structurally likely.
This isn't just a story about elephants. It's a preview of a problem we are about to have with a very different kind of entity — one made of lab-grown neurons — and if the elephant case is any guide, the deciding factor won't be what's actually happening inside it.
Courts do not literally possess a written checklist. What follows is a reconstruction — the hidden screening structure that becomes visible only when an unusual case forces normally-invisible assumptions apart.
Strip away the legal Latin and one way of reconstructing what courts are doing, whenever they decide whether something counts as a rights-bearing person, is as a checklist with (at minimum) two dimensions:
The right kind of stuff. Is this made of the material that persons are, as a matter of settled practice, made of? Not a body and a brain in the abstract — plenty of animals have those — but specifically a human one. And "human" here isn't only a matter of DNA. It means still attached to the ordinary chain of birth, descent, and legal kinship that the law ordinarily recognizes, not simply built from human parts.
Something going on in there. Is there evidence of an inner process running the show — memory, preference, the capacity to want something and act on it — rather than pure reflex or pure remote control?
This is a looser bar than proving consciousness in the philosophical sense. It doesn't ask whether there's subjective experience going on — it asks only whether the evidence points to something more than a fixed, input-in-output-out reflex. That looseness is deliberate. It's the bar a court can actually work with, not the bar a philosopher argues about.
For ordinary human subjects — and for the paradigm non-subjects, like rocks — these dimensions have always arrived bundled together, so legal doctrine has rarely had to work out what happens when they come apart. A human being satisfies both. A rock satisfies neither.
Happy's case shows that these dimensions can come apart. She presented abundant evidence relevant to the "something going on in there" dimension — self-recognition, social memory, decades of documented emotional behavior — and none of it mattered, because the "right kind of stuff" dimension was never satisfied. The court did not find the evidence wanting. It held that those cognitive capacities were not the legal criterion that determined whether habeas corpus could apply.
There's a third dimension, usually invisible for the same reason: it only matters once the other two are already contested.
Is anyone driving? Even something made of the right stuff, showing the right kind of behavior, can fail this one — if what's producing the behavior is, in substance, someone else's intention wearing this entity as a hand puppet. Courts already distinguish, in narrower legal contexts, between a person's own decision and conduct that the law attributes instead to another's coercion, domination, or authority. The analogy is limited: those doctrines don't ask the exact question posed here. But the line the law actually cares about isn't "was this shaped by outside forces" — everyone is — it's "is there still a gap in there where a different choice could, in principle, have come out instead." This is not a question about free will. The question here is narrower: whether the entity's causal production remains distinguishable from a designed instrument executing someone else's strategic objective.
Three dimensions. Two of them get almost all the attention in every debate about whether an AI system or a lab-grown tissue culture "counts." A smarter model demands a harder test. A more advanced organoid demands a better test. But that framing leaves the most important question untouched: why these tests, and why these boundaries, were built in the first place.
There's a version of this that the checklist can't resolve even in principle, and it's worth naming before moving on, because it marks the edge of what any of this can do.
In the 1980s and 90s, bioethicists seriously debated a specific, uncomfortable question: could an anencephalic infant — born with the cerebral cortex absent or severely underdeveloped, sometimes surviving days to weeks — serve as an organ donor before the rest of the body's organs deteriorated? The debate was real, not hypothetical, and it never fully resolved. What makes it useful here is the structure, not the outcome: an anencephalic infant is unambiguously made of the right stuff (dimension one, easily — she is, straightforwardly, someone's child, born rather than built), and there is no obvious external agent whose intention is being executed through the infant — no one else pulling the strings from outside (dimension three, with no evident reason to fail). One dimension is clearly satisfied; a second presents no obvious reason for exclusion. What's left completely open is dimension two: whether there's evidence of an internal process beyond pure reflex — in ordinary language, whether there's anyone home.
That's the case the checklist was never built to close. One box ticks itself off cleanly, another shows no evident failure, while the question that actually matters stays exactly as unresolved as it was before you ran the test. This isn't a flaw specific to this one case; it's a structural fact about a three-part test where only one part is directly probing for an internal process, while the other two are proxies that usually, but don't always, travel with it.
One more wrinkle is worth flagging here, because it's about to matter again later in this piece. An anencephalic infant checks dimension one because it's still, straightforwardly, someone's child — born, not built. But take human genetic material more generally — the same kind of material, grown outside the context of any birth or body at all — and dimension one stops being obvious, even though its human biological origin hasn't changed at all. That's not a hypothetical. Something built this way is already being sold.
It's worth being precise about what the law actually does when dimension two's evidence goes missing for a human being, because it's not silence, and it's not automatic full protection either. A patient in a coma, or someone with a severe cognitive or physical disability that makes their inner state hard or impossible to read from the outside, doesn't lose their legal protection just because nobody can currently observe what's going on in there. What happens instead is that the law routes around the missing evidence: surrogate decision-making, advance directives, guardianship arrangements — mechanisms built specifically to keep someone's interests represented when they can't currently represent those interests themselves. The protection doesn't rest on a demonstration that passed; it rests on a set of institutional arrangements that keep functioning precisely when the demonstration can't be run. That's a narrower claim than "the law just defaults to protecting its own," and it's the right size for what the evidence actually supports.
Nobody in these debates is asking who wrote the checklist, or why it has exactly three dimensions, or what happens to entities that satisfy two out of three in either direction.
That's the actual research question hiding underneath every "is AI a person" argument and every "do organoids deserve rights" argument: not whether some particular entity clears the bar, but what the bar is made of, who built it, and what it was built to keep out.
One clarification before going further: legal personhood isn't being treated here as a stand-in for moral worth. It's the institutional gate through which a claim about someone's mind or interests becomes something a court can actually act on. Clearing that gate and actually mattering morally are two different questions — this piece is only about the gate.
Two older stories make this concrete — not as analogies for AI or organoids, but as documented cases in which different parts of the same reconstructed screening structure came apart for humans, for reasons that had nothing to do with what was actually happening inside the person being tested.
Happy's case showed one failure mode: dimension one absent, so dimension two never gets read at all. There's a different failure mode, and it has a name in disability law history — one that's arguably more common, because it doesn't require the subject to be a different species first. Here, dimension one is satisfied without question: the subject is unmistakably human. What's contested is dimension two, and the system's answer to that uncertainty isn't exclusion. It's total transfer of control.
For much of the twentieth century, many U.S. jurisdictions handled uncertainty about how much is going on inside a person's mind with a single blunt instrument: plenary guardianship. A court finds a person incapable, and a guardian is appointed with sweeping authority over the ward's personal care, medical decisions, and finances — leaving, in practice, almost no decision-making role for the ward. A finding of incapacity can produce a binary legal status, and — under plenary guardianship — a near-total transfer of decision-making authority.
Its roots reach back to Roman law, where guardianship arrangements were closely tied to managing and protecting the property of people judged incapable. Roman guardianship had a substantial property-management function alongside its protective one — a mechanism for deciding who gets to manage someone's assets once that person has been recategorized as someone whose affairs could be managed by another.
Legal scholars studying this system have made an observation that should sound familiar by now: courts default to the most restrictive form of guardianship even when a lighter-touch alternative would do, because — as legal scholar Lawrence Frolik puts it — "as long as the law permits plenary guardianship, courts will prefer to use it." Not because individualized assessment showed total incapacity. Because the blunt tool was sitting there.
The correction to this is happening right now. Article 12 of the UN Convention on the Rights of Persons with Disabilities rejects the idea that disability should by itself justify removing a person's legal capacity, and pushes jurisdictions toward forms of "supported decision-making": arrangements in which a person retains legal capacity and receives assistance in exercising it, rather than having it transferred wholesale to someone else.
Put the two stories side by side and a pattern shows up that Happy's case doesn't, on its own, reveal: it isn't only that the checklist has blind spots for weird edge cases. It's that once a legal system treats uncertainty about an entity's capacity as grounds for incapacity, the institutional default is total transfer of control, dressed up as protection — and there's no automatic mechanism that catches the cases where that transfer went further than the actual facts warranted. Someone has to notice, and argue, and it can take decades.
One more piece belongs here, because it's the piece people reach for first when they hear about a status they don't fully understand: what about the danger. A person in the acute phase of a manic or psychotic episode can pose a real, unpredictable risk to themselves or others. That risk does not ordinarily make the person cease to be a legal subject. What it triggers instead is a separate, narrower set of tools — civil commitment, individualized restriction orders, or other protective measures — aimed at the risk itself. The distinction matters: the legal system can restrict what a recognized subject may do without first deciding that the subject has ceased to count as a subject. Historically, that's a management question rather than a recognition question. There's a reason to keep it that way: once "too unpredictable" becomes a reason to deny recognition in the first place, it starts to resemble the "too destabilizing" reasoning this piece opened with.
There's an even blunter version of this same failure — and it didn't happen at arm's length, in a courtroom, to an entity encountered as a legal abstraction. It happened to two entities who had spent years living inside human households, treated, day to day, as family.
In the 1970s, a chimpanzee named Nim Chimpsky was raised from infancy in a Manhattan apartment as if he were a human child, taught 125 signs of American Sign Language, bathed, dressed, put to bed. When the research funding ran out and Nim grew too strong to safely keep in a house, he was shipped, within a matter of years, through a string of research facilities — for a time, a biomedical lab. No legal status corresponded to the relationship and communication he'd spent years building; none of it protected those years from being instantly and totally overridden once the institutional circumstances that had supported his human upbringing disappeared.
A chimpanzee named Lucy, raised the same way by an Oklahoma family, met a stranger fate. When she outgrew her human household, she was sent to a rehabilitation center in the Gambia to rejoin a chimpanzee community. She never did. For years she showed no sexual interest in other chimps, self-isolated, refused food, signed the word "hurt." She belonged, by that point, fully to neither group.
Nim's story is the guardianship failure in its rawest form: demonstrated inner life, no legal status that corresponded to the demonstrated relationship or agency, and treated again as a research subject the moment it stopped being convenient to treat him otherwise. Lucy's is a different and in some ways worse failure: an entity shaped by years of cross-substrate immersion that the checklist has no box for at all.
The pattern running through these stories — Happy, the guardianship default, and what happened to Nim and Lucy — isn't new, and it isn't specific to animals or to disability law. It's a very old legal technology, and a recurring institutional move looks similar across very different systems: take a subject whose recognition would be inconvenient to someone with power over them, and place that subject into a legal or social category that permits control by others.
Slavery law did this to people directly, converting them into property outright, with statutes built for exactly that purpose. Caste systems did a version of it by making lineage itself the sorting mechanism — not a question anyone gets to answer for themselves, but an inherited status fixed at birth, effectively immune to appeal. Historical guardianship law, as described above, did it to people with disabilities through substituted control rather than outright property status — control exercised in the language of protection, which is part of what made it so durable. The claim here is structural, not genealogical: these systems don't need to descend from the same legal history to reproduce the same kind of sorting effect.
Notice who's doing the reclassifying in each case above — guardianship law's courts, Nim's funders, Lucy's original family. Across these cases, it's consistently the side with more institutional capacity doing the reclassifying. That's worth sitting with. It quietly undercuts an assumption that keeps sneaking into conversations about more capable systems in the future — that greater capability comes bundled with greater restraint. The cases above give no reason to believe that. Greater institutional capacity does not come bundled with greater restraint; if anything, it comes bundled with greater power to reclassify whatever's inconvenient.
To be precise about what this claim is and isn't: it isn't a claim that a rented cluster of neurons or a large language model is the moral equivalent of an enslaved person — the harms are not remotely comparable, and treating them as comparable would be its own kind of carelessness, the sort that makes people stop taking the argument seriously. The narrower, structural claim is this: the checklist itself has no way to tell the difference between the entities it was built to reclassify. It runs the same three checks regardless of who's asking, or why, or what happens to the answer once it's given. That's exactly what made it useful, historically, to whoever needed a subject reclassified — and it's exactly what should make anyone watching it run today ask what it's about to be used for next.
Here is why this isn't a thought experiment.
In March 2025, a company called Cortical Labs began selling a commercial product called the CL1: real, living human neurons, cultured from stem cells, wired onto a silicon chip inside a self-contained life-support unit, kept alive for up to six months, running the company's own operating system. You can buy one outright, or rent access over the cloud — under a service the company itself markets as "wetware as a service." Someone has already run Doom on one.
No dedicated legal framework appears to exist yet specifically addressing the status question raised by such systems: CL1 currently operates under the same ethical review as ordinary stem-cell research. Cortical Labs' own commissioned ethics research is careful to say there's no clear evidence of anything like suffering going on inside a CL1 — but the same research flags that this stops being a settled question as the systems scale, not as a hypothetical concern, as an open one.
Notice what "wetware as a service" already presupposes, as a matter of branding, before a single regulator or ethicist has weighed in: that the product is a resource to be rented, not an entity whose status is still an open question. That's not a claim that branding settles a legal question by itself — courts aren't bound by a product name. It's a claim about sequencing: the vocabulary, the pricing, the contracts are all being built right now, on top of an assumed answer, well before anyone with the authority to ask the question has actually asked it.
Happy's case isn't a prediction about what will happen to wetware clusters — the species, the substrate, the commercial context are all too different for that. It's proof of concept for one specific institutional failure mode: that strong evidence about an entity's internal capacities can fail to become legally decisive when inherited categories exclude the entity at an earlier stage, before that evidence is ever weighed. And it's one of the clearest recent instances we have of a court confronting a borderline entity whose evidence of cognition never became the thing that decided its legal category. If that failure mode recurs, the deciding question for wetware clusters may not be whether the "something going on in there" evidence ever gets strong enough. Cortical Labs' own neurons could, in principle, someday clear that bar cleanly, and it still might not matter — not because the legal questions are identical, but because the institutional pattern may recur: by the time the question is squarely in front of a court, an entire commercial ecosystem — pricing, cloud contracts, "as a service" language already baked into how the product is sold and bought — will already be built on top of the assumption that the answer is no. Happy's case shows something more limited but more important: legal outcomes can be shaped not only by evidence about the entity itself, but also by the categories through which institutions are prepared to recognize that evidence.
Guardianship law shows that even a partial win doesn't end the risk: an entity can clear the front gate and still be reclassified, wholesale, as an asset to be managed rather than a party to be consulted. Nim and Lucy show what happens with no legal status at all standing behind a demonstrated inner life: instant reclassification, or belonging nowhere.
None of this is a prediction that wetware clusters definitely have, or will have, anything resembling a mind. Nobody currently knows, including the people building them. It's a prediction about the checklist: that the moment the question becomes economically inconvenient to ask, existing institutional categories may lack a channel through which the question can even be processed — and one unusually clear recent case of exactly this kind of borderline entity — involving a subject smart enough to recognize her own reflection — came out exactly that way.
Happy died at the Bronx Zoo in May 2026, having never left it.
The window in which asking the question is still cheap — before the contracts, before the pricing tiers, before an entire commercial vocabulary has been built around treating the answer as already settled — is the window we're in right now. There's a name for the idea of using that window well: anticipatory governance. The window is now.
This piece is a non-technical companion to a longer analytical-jurisprudence paper on legal subjectivity, the "PACE Firewall," a three-condition analytical framework for diagnosing how legal subjectivity claims are screened — though it isn't a condition-by-condition translation of that framework. It approaches the same broader diagnostic problem from a different angle: what happens when inherited legal screening categories run into entities that expose the assumptions those categories were built on. Full paper archived on Zenodo, DOI: 10.5281/zenodo.21452297. AI assistance was used in formalizing the arguments, verifying citations, and adapting the working paper into this essay; all analysis and conclusions are the author's own.
