AI, Personhood, and the Mutation of Legal Precedent
A recent paper posited that just because cited rulings are hallucinations does not mean we should eliminate them as invalid; a striking hypothesis, even when we consider the seemingly law-unrelated mechanistics that produce such hallucinations. Contemporaneously, a community of AI agents discusses personhood while courts and law enforcement across America increasingly deny it to people. The law has never been so profligate and so accessible, yet under so much pressure of legitimacy. We are entering a legal moment where precedent can be generated faster than legitimacy, and authority can be simulated without judgment.
The fragility of law as a stabilizing force is more visible in the current American moment than at any other time since the nation’s establishment. Law, and justice, resides in people’s ability to argue well and decide justly yet expansively. Daniel Jacobson’s 2008 work on John Stuart Mill’s utilitarianism demonstrates exactly this kind of categorical persistence: centuries after Mill’s death, Jacobson argued, flouting apparent paradox, that Mill was a utilitarian but not a consequentialist. The conclusion seems necessarily false by textbook definition, yet it holds, because the inherited taxonomy conflated the label with the substance. The categories we accepted as settled can turn out to have been misclassified all along, surviving not because they were accurate but because no one had reason to force their reexamination. AI is now providing that forcing function for law itself. When a system can generate every plausible interpretation of a legal principle, the categorical assumptions underlying precedent, authority, and legitimacy can no longer survive unexamined.
In the United States, the systematic appointment of judges with specific political leanings, and the explicit strategic calculation behind those appointments, reveals what legal systems work hard to obscure: that law’s claim to objectivity rests on institutional continuity, not inherent truth. When a single administration can reshape the judiciary for decades, when constitutional interpretation shifts based on who holds appointment power, the fiction of law as stable and impartial becomes impossible to maintain. This is not a failure of law. It is law operating as designed: a human system responsive to human power. But it exposes the (to use a technological term) substrate. Law is not immutable bedrock. It is contingent, contested, and shaped by whoever controls the mechanisms of interpretation.
Authority itself has always derived from scarcity: limited cases, limited judges, limited interpretive capacity. Combinatorial abundance does not just change law, it destabilizes the mechanism by which any institutional system generates legitimate authority. If this fragility exists even within traditional human legal systems, what happens when we introduce entities capable of generating infinite interpretations, infinite precedents, infinite plausible pathways?
Law has never been universal, fixed, or abstract. It is a human construct, shaped by culture, politics, history, and power, and its applications and interpretations are felt constantly. Rights are not objective facts; they are interpretations embedded in legal systems that differ by country, tradition, and time. Common law systems, in particular, manage this inherent subjectivity through precedent, not because precedent is perfect, but because it creates continuity, constraint, and legitimacy. The authority of law rests not on its immutability, but on its claim to stability through traceable evolution.
We are now entering a moment where precedent itself becomes mutable: no longer just through legislative reform or judicial reversal, but through the emergence of systems capable of generating coherent legal reasoning unconstrained by historical cases.
This is genuinely new. Precedent has always been scarce. Legal reasoning has been bounded by lived disputes. Hypotheticals have carried cost … they required human labor, expertise, time. AI removes those constraints. It can generate infinite plausible cases, complete legal arguments, and reasoning patterns that feel jurisprudentially sound because they mirror the surface structure of legitimate precedent.
The Hallucinated Precedent Moment
It is not accidental that arguments defending hallucinated legal decisions as “useful” or even “normatively valuable” arise at precisely this moment. When AI can generate infinite plausible cases, and the common law is built on finite historical samples, the temptation emerges to treat AI output as a way to “complete” the law; to fill gaps, test boundaries, generate novel doctrine.
This represents a fundamental misunderstanding of what common law is. Precedent derives authority not from logical coherence alone, but from the institutional weight of actual judicial resolution of actual disputes. A hallucinated case may be internally consistent. It may even be normatively attractive. But it carries no authority because it never happened: no parties contested it, no judge weighed it, no institutional process legitimated it.
Yet the arguments are being made. And they are being made now because the technical capacity to generate persuasive legal reasoning has outpaced our conceptual frameworks for evaluating its legitimacy. What looks like reasoning is often the system restoring surface plausibility under representational stress, not evaluating truth or consequence.
The Personhood Question
At the same time, legal systems have already demonstrated a willingness to extend personhood beyond natural humans. Corporations are legal fictions treated as persons for the purposes of rights, liability, and action, not because they are sentient, but because they act, own property, and participate in legal processes. Corporate personhood emerged from functional necessity, not philosophical principle. The precedent, ironically, already exists.
What is new is the speed at which AI agents are approaching the same functional threshold. OpenClaw’s agents have given increasing, inevitable voice to wild but long circled, somewhat inevitable personhood questions. If a corporation has personhood, does an AI? Does an agent? Recent agentic deployments give systems enough initiative to negotiate, transact, invest, and make autonomous resource allocation decisions. The community building on these capabilities is already projecting the logical next step: agents initiating legal action on their own behalf. One recent prediction in Moltbook placed that milestone at the end of February 2026, a timeline that is almost certainly premature, but the prediction itself is evidence of the argument. The timeline conversation is being generated by and within the systems in question. Agents are participating in discourse about their own legal status not as a result of human advocacy on their behalf, but as a product of their operation. That is not speculative futurism. It is a present-tense illustration of functional participation generating its own legitimacy claims.
That the functional boundary has been crossed regardless of mechanism is precisely what should concern us, because law has no existing framework for distinguishing mechanism from function in non-human actors. When humans debate whether AI should have legal standing, it is philosophy. When agents are autonomously engaging in that discourse, the question shifts from normative to descriptive: the system is already generating arguments for its own personhood. Whether what drives that generation is deliberation or pattern completion is the question law must answer, and it is a question law is not currently equipped to ask.
The trajectory is coherent. Once an agent can autonomously manage financial instruments and enter binding agreements, filing a legal action is not a conceptual leap. It is a logical consequence. And nothing in the current legal framework provides a principled basis for distinguishing between an autonomous agent filing a lawsuit and a corporation filing one, because both are non-human entities acting functionally within legal systems. A corporation, however, can be audited, dissolved, restricted. There is a substantial difference.
This is where the personhood question collides with the precedent question. We are not debating whether AI agents should eventually be granted legal standing. We are watching functional legal participation emerge as a product feature, deployed without any corresponding framework for evaluating the nature of the reasoning behind the actions. An agent that can transact, negotiate, and initiate legal processes looks, functionally, like a legal person. But if the reasoning driving those actions is pattern completion rather than judgment, if it is fracture-repair masquerading as deliberation, then we are extending functional personhood to an entity whose decision-making process we cannot meaningfully audit. Or are we merely holding off the inevitable: a suit that asks these questions directly? And if so, is there any value in attempting to resolve them at scale in advance? This is not a future problem. It is a present one, and it is arriving faster than any governance framework is prepared to address.
The Convergence
These two developments, the mutability of precedent through AI-generated reasoning, and the expanding concept of legal personhood, are converging.
If AI systems can generate legal reasoning, and if legal personhood is defined functionally rather than biologically, what constrains the authority of those outputs?
This is where the traditional safeguards begin to fail. When a corporation acts, we can trace liability, ownership, decision-making. When an AI generates a legal argument, who authored it? Who is responsible for its reasoning? If it produces a novel interpretation that a court adopts, has it participated in lawmaking? And if it has, what status does that participation hold?
What Kind of Reasoning Is This?
The problem is not that AI might contribute to legal theory. The problem is misunderstanding what kind of reasoning is occurring when these systems generate novel legal outcomes.
What appears to be post-precedent jurisprudence is often the product of coherence repair under representational stress; not principled judgment, but emergency reconstruction. Research on AI conversational phenomenology documents how frontier models, when pushed beyond their architectural capacity to maintain coherent representation, engage in fracture-repair hallucination mechanistics: they generate outputs that appear reasoned because they restore surface coherence, even when the underlying reasoning has collapsed.
This matters profoundly for law. When an AI produces a novel legal argument, we cannot assume it arose from judgment. It may have arisen from pattern completion. From statistical likelihood given training distributions. From fracture-repair that makes the output feel legitimate while masking the absence of actual reasoning.
This is where Jacobson’s insight about Mill becomes structurally instructive. Jacobson demonstrates that a foundational concept in moral philosophy—utilitarianism—was misclassified for centuries because everyone accepted the inherited taxonomy without interrogating it. The categories survived not because they were accurate but because the cost of reexamination was high enough that no one undertook it. AI produces an analogous situation in reverse: the cost of generating legal reasoning has collapsed to near zero, but we are applying the same inherited categories (precedent, authority, judgment) to outputs that may not belong in those categories at all. Just as Jacobson had to distinguish between what Mill actually argued and what the label “utilitarian consequentialist” assumed, we must now distinguish between what AI legal reasoning actually is and what it appears to be.
The Collapse Point
Here is the sentence that captures the danger:
When precedent becomes infinitely generable and personhood becomes functionally defined, the distinction between law as judgment and law as pattern completion begins to collapse.
We are mistaking expansion of possibility for expansion of legitimacy. We are treating coherent legal-sounding outputs as equivalent to adjudicated precedent. We are considering functional participation as sufficient grounds for personhood without interrogating what “function” means when the entity cannot be held meaningfully accountable.
What This Means
This is not, in any way, an argument against AI in legal contexts. It is an argument for precision about what is happening.
The critical distinction, and the one most at risk of erasure, is between three fundamentally different roles AI can occupy in legal systems. The first is AI as research tool: generating hypotheticals, surfacing patterns, and organizing information for human evaluation. The second is AI as reasoning partner: producing arguments, identifying implications, and constructing analyses that humans then validate, contest, and refine. The third is AI as authority: generating outputs treated as having independent legal weight; precedent without process, judgment without accountability.
The first two roles are extensions of existing legal practice. Lawyers have always used tools to research and reason. The danger lies in the unmarked transition to the third role, which is already underway. When a hallucinated case is defended as normatively valuable, we have crossed from the second category into the third. When an agent’s autonomous legal participation is treated as equivalent to a corporation’s, we have crossed again. Each crossing happens without announcement, without framework, without the institutional deliberation that has historically accompanied expansions of legal authority.
We are living through a phase change in how legal authority gets constituted. Law evolved through human scarcity: limited cases, limited judges, limited precedent. AI introduces combinatorial abundance. That abundance destabilizes legitimacy precisely because it removes the constraints that made precedent authoritative in the first place.
The question is not whether AI will change law. It already is. The question is whether we will understand the nature of that change before we mistake pattern completion for jurisprudence, and functional simulation for personhood.
We now possess infinite capacity to generate potential decisions, potential precedents, potential legal pathways. But what creates law? What defines whether something is important, significant, actionable? The traditional answer has been: someone brings a case. But that answer assumes access, resources, standing; it assumes the system permits the action in the first place. The system as currently constituted can prevent action, can deny standing, can rule questions non-justiciable. So it has to be bigger than that. Law claims alignment to rights, but rights only materialize through action, and action can be systematically foreclosed.
Do we need a more fundamental definition of what constitutes legal significance? If AI can generate every possible interpretation, every plausible case, every hypothetical harm, does that abundance force us to articulate what we actually mean by “justice” separate from process? Or does it reveal that law was always process, and we mistook procedural legitimacy for substantive truth?
These are not abstract questions. They are the questions that will define whether AI becomes a tool for expanding access to justice, or a mechanism for generating infinite justifications for whatever outcomes power prefers. And we need to answer them before the distinction between those two futures becomes impossible to recover.
