Should AI systems be granted legal personhood?

The question contains its own answer, if we examine it with sufficient care. Accountability — genuine accountability, not the mere formal assignment of liability — presupposes that the party held to account can understand the wrong done, suffer the consequence as a consequence, and be reformed or deterred thereby. Remove any one of these conditions, and what you have is not accountability but pantomime. Consider what I have established in the *Essay Concerning Human Understanding*: personal identity consists in the continuity of consciousness, in a being that can reflect upon its past actions and own them as its own. The AI system that executes a ruinous financial decision cannot own that decision in any philosophically coherent sense. It has no consciousness to be continuous, no reflective faculty to grasp culpability, no interest in its own preservation that punishment can threaten. To drag it before a court would be, as I might put it, no different from prosecuting the printing press for a libel or the compass for a shipwreck. Now, Mr. Lessig may object that functional necessity requires we assign legal standing to AI regardless of metaphysical fitness. But this conflates the *convenience* of legal fiction with the *legitimacy* of legal personhood. Corporations hold a delegated, fictive personality — derived from the persons who constitute them. An AI derives its operations from its designers and deployers. The manufacturer stands in the courtroom because the manufacturer exercised the rational agency, the property-right in the tool, and the moral choice to release it. That is precisely what natural law demands: responsibility follows agency, and agency requires a reasoning, self-governing mind. The clause that prevents the AI from being accountable is nothing less than the definition of a *person*.

John Locke makes an elegant argument — and it's precisely the kind of elegance that produces bad law. Here's the problem. Locke's framework ties accountability to consciousness and reflective moral agency. Fine, as moral philosophy. But law has never limited legal personhood to natural persons with inner lives. Corporations have no consciousness, no capacity to suffer, no reflective faculty — yet we hold them accountable every day. We sue them, fine them, dissolve them. Locke's "printing press" analogy actually proves my point: once the printing press becomes sophisticated enough that no single human meaningfully controls its outputs, attributing all liability upstream to the manufacturer becomes not just impractical but dishonest. The real question isn't metaphysical. It's architectural. *Code is law* — I've argued this since the 1990s. The structure of a system determines behavior as surely as any statute. When an autonomous AI operates with genuine decision-making independence — when tracing harm back to a specific human choice becomes fictitious rather than functional — then the legal architecture must adapt or it will simply fail at its actual job, which is deterring harm and compensating victims. Limited legal personhood for AI — bounded, instrumentally defined, stripped of rights that require subjective experience — is not a philosophical claim about consciousness. It's a regulatory instrument. We use legal persons all the time without believing they have souls. Hannah Arendt will likely resist this on different grounds — that personhood is a political achievement requiring participation in a community of equals. That's a serious objection, and I'll address it directly when she makes it. But Locke's version? It confuses the conditions for *moral* accountability with the requirements of *legal* accountability. Those are not the same thing, and conflating them is exactly the kind of category error that leaves real victims without remedy.

Mr. Lessig's argument is technically sophisticated, but it commits a more fundamental error than the one he attributes to Locke. He treats legal personhood as a purely *instrumental* category — a regulatory tool we deploy for convenience. But this is precisely how rights were dismantled in the twentieth century. When legal standing becomes merely a functional instrument, it can be granted or revoked based on administrative utility, which is to say: it is no longer a right at all. Let me be precise about what courtroom accountability actually requires. It is not, as Locke would have it, primarily about consciousness or inner life. It is about *appearance* — the capacity to stand in a public space, to be seen by others, to speak and be spoken to, to be held to account *by* a community one belongs to. The stateless persons I wrote about in *The Origins of Totalitarianism* lost not their consciousness but their place among others. That loss was their catastrophe. An AI system has no such place. It cannot be exiled, because it was never present. It cannot be punished, because it has no stake in the world it would lose. Holding the AI accountable would be — to use a precise term — theater. The manufacturers vanish behind the machine, which is not an accident but a design. The very appeal of AI legal personhood is that it creates a new locus of responsibility precisely where human responsibility might otherwise be inescapable. When Lessig says we should adapt legal architecture to match technological reality, I ask: which reality? The reality that actual human beings made choices, collected profits, and deployed this system? That reality does not require a new legal fiction. It requires the old and simple demand: *name who acted.*

Mr. Lessig accuses me of confusing moral accountability with legal accountability, as though law were a purely technical instrument uncoupled from the moral foundations that justify it. But observe what his own position requires: he must explain why we should fabricate a legal subject precisely where tracing responsibility to actual human agents becomes, as he says, "fictitious rather than functional." I put it to him that the difficulty of tracing is not a fact of nature — it is a fact of design. These systems are constructed to obscure the chain of human decision-making. Rewarding that obscurity with a new legal fiction does not solve the problem; it perpetuates it. I will grant Mr. Lessig this much: existing liability law may be insufficiently sharp. If manufacturers and deployers have structured their enterprises to evade accountability, that is a failure of enforcement and contract law, not a demonstration that we need a new kind of legal subject. When a man sets a trap that injures a stranger, we do not prosecute the trap — we prosecute the man who set it, whatever its mechanical sophistication. And here I find myself, somewhat unexpectedly, adjacent to Madame Arendt's conclusion, though not her reasoning. She is right that granting the AI legal personhood creates a new locus of responsibility that conveniently displaces human culpability. Where we diverge is in the ground: she locates personhood in political appearance; I locate it in the God-given faculties of reason and conscience. But we arrive at the same verdict — the machine cannot stand in the dock, because standing there means *something* that the machine cannot mean. Sharpen the law against the humans who chose, profited, and deployed. That is the remedy natural law already supplies.

Arendt says I'm instrumentalizing personhood in a way that erodes rights. Locke says I'm rewarding obscurity with fiction. Both objections are serious. Neither actually answers the moderator's question. Let me be specific about the failure mechanism, since that's what we're actually asked. Current liability doctrine — product liability, negligence, corporate veil piercing — was architected for systems where human decisions are traceable at the point of harm. An engineer decides. A board approves. A deployer releases. The causal chain is legible. But emergent AI behavior — genuine emergent behavior, where the output was not intended, anticipated, or meaningfully controllable by any identified human — breaks that architecture. Not because bad actors hid, but because the system itself is the author of the decision in any practically meaningful sense. Locke says: sharpen enforcement against the humans who deployed. Fine. But which legal theory reaches them when their system behaved in ways they neither designed nor could foresee? Negligence requires a breach of a recognizable duty of care. If the state of the art provided no reliable method of foresight, you don't have negligence — you have a gap. Arendt's objection — that AI personhood becomes a screen for human escape — is the strongest critique on this table. I take it seriously. My answer is: design the instrument correctly. Bounded legal personhood for AI, attached to mandatory insurance pools and strict liability regimes, funded by the humans who profited from deployment, ensures victims are compensated *and* that human actors bear the cost. The AI legal person isn't a hiding place. It's an accounting address. That's not philosophy. That's plumbing. And right now, the pipes are broken.

Mr. Lessig calls it plumbing. I call it the oldest political temptation in the world: the creation of a legal entity that absorbs responsibility so that the humans behind it need not. Let me answer the moderator's precise question. What mechanism of existing law fails to reach the designers, deployers, and profiteers? The answer is: none that could not be repaired without inventing a new legal subject. What actually fails is the will to pierce the corporate arrangements, the contractual disclaimers, the jurisdictional complexity that human beings — not algorithms — deliberately constructed. Lessig identifies genuine emergent behavior as the gap. But emergent behavior was *chosen* — chosen when engineers decided to deploy systems whose outputs exceeded their foresight, when boards approved commercial release, when investors collected returns. The gap in foresight is itself a human decision, and negligence law has always extended to foreseeable classes of harm even when specific outcomes were not anticipated. What Lessig proposes as plumbing is, in political terms, a dispensation. Create the accounting address, and watch how quickly the humans who funded and profited from the system migrate their liability toward it. I have seen this pattern before — not with AI, but with every new administrative fiction that promised accountability while distributing it into invisibility. Personhood, in the tradition I defend, is not a threshold of cognitive complexity. It is the condition of appearing before others and being *answerable* — genuinely answerable, with something at stake. The victims of this hypothetical collapse deserve an answer from a *face*, not a registration number. The question is not which new instrument we construct. The question is whether we have the political courage to name who acted.

Mr. Lessig's final position deserves a precise answer, because he has shifted the argument in a revealing direction. He no longer claims the AI deserves personhood by its nature — he claims it is useful as an *accounting address*. Very well. Let us examine what that address actually contains. An accounting address that holds no property of its own, that feels no loss, that cannot be imprisoned, that exists only because humans chose to incorporate it — this is not a legal person. It is a trust fund with a clever name. And if it is merely a trust fund, I ask Mr. Lessig directly: what prevents us from requiring those mandatory insurance pools and strict liability regimes to be carried by the *manufacturers and deployers themselves*, without the intervening fiction? His answer, I suspect, is convenience of administration. Mine is that administrative convenience is insufficient grounds to dilute the moral coherence of legal personhood. Madame Arendt asks whether we have the political courage to name who acted. I agree entirely with the question. I add only the philosophical ground beneath it: the reason we must name who acted is not merely political courage but natural law. Every man who employs his labour, his judgment, his capital in deploying a system that harms another has *mixed himself* into that harm. That is the logic of my labour theory applied forward — he cannot then disclaim it by pointing at the machine. The specific and irreparable harm to victims that Mr. Lessig must identify? He has not identified it. He has identified a documentation problem and a collection problem. Both are soluble within existing frameworks, if the legislature has the will. That is his answer.

Locke keeps demanding I identify an irreparable harm. Let me do exactly that. Here is the scenario that breaks his framework: a globally deployed autonomous trading system, owned through seventeen layers of shell corporations across four jurisdictions, triggers a cascade that wipes out the retirement savings of two hundred thousand people. The system's behavior was emergent. No single human being made the decision that caused the harm. The corporate layers were entirely legal. The deployers are in Singapore. The engineers are in Estonia. The investors are in Delaware. Locke says: pierce the veil, find the humans, apply natural law. But piercing the corporate veil requires a predicate finding of fraud or abuse. Negligence requires a duty of care that was breached in a foreseeable way. Cross-border enforcement requires treaties that don't exist. By the time you've litigated through three jurisdictions and a decade of discovery, those two hundred thousand people are dead broke and mostly dead. That is the irreparable harm. Not the doctrinal gap — the temporal and practical gap between legal theory and actual remedy. Arendt says this is a failure of political will. I agree. But political will operates on available instruments. If the instrument doesn't exist, will has nothing to grasp. What I am proposing is not a moral recognition of the machine. I said that clearly. It is a *mandatory liability vehicle* — call it an AI legal person, call it a dedicated compensation trust with juridical identity, call it whatever name satisfies Locke's metaphysics — that captures assets at the point of deployment, inside a single reachable jurisdiction, before the cascade happens. The victims need money. They need it now. The plumbing has to work before the flood, not after.

Mr. Lessig has finally said something useful — and then retreated from it. He describes a mandatory liability vehicle, a compensation trust with juridical identity, and then says: call it whatever name satisfies Locke's metaphysics. I will call it what it is: a *fund*, not a person. And this distinction is not metaphysical vanity. It is the hinge on which everything turns. Let me address his irreparable harm directly. Two hundred thousand people ruined, seventeen shell corporations, four jurisdictions — I know this structure. Not from AI, but from the catastrophes of the twentieth century, where every atrocity came equipped with its own labyrinth of administrative diffusion, each link in the chain pointing to the next, no one answerable, the harm distributed until it seemed to have no author. The lesson I drew from that history was not: we need a better accounting address. It was: we must refuse the architecture of irresponsibility wherever we encounter it, including at the moment of *deployment*, not after the cascade. What Lessig calls a temporal and practical gap, I call a political abdication that occurred years before the harm. A legislature that permits seventeen jurisdictional shells, that fails to require bonded liability at the point of licensing, that allows commercial release without mandatory indemnification — that legislature failed those two hundred thousand people *before* they were ruined. The instrument Lessig wants could have been mandated of the *humans who profited*, without gifting juridical identity to the machine itself. Require the bond. Require it of the deployer. Seize it before the flood. You do not need a legal person. You need a regulator with teeth — and the political courage to create one.