Autonomous Organizations and the Decline of Anthropocentric Law

Abstract

This article charts the legal and technological developments that have enabled nonhuman systems, such as artificially intelligent software, to take actions that have significant consequences under private law, such as the making of contracts, the management of companies, and the incursion of tort or restitutionary liability. Because of these developments, it is important to recognize that while the law still serves human ends, “participants” in the legal system are no longer exclusively human; developments in organizational law have driven a shift so that, in a meaningful sense, nonhuman systems are more than the direct instrumentalities of the human beings who have set them in motion. The article gives an overview of the relevant legal and technological developments, evaluates pressures that they may put on the doctrines and concepts of private law, and considers their broader future possible effects on legal theory and on the goals that commentators suppose that law is to adopt.

1. A Short History of the Decline of Anthropocentric Law

Until the rise of automation, the law rarely responded to nonhuman activity. To be sure, the law traditionally had concepts of “acts of God” and “greater forces” outside the control of human beings (Corbin 2025)—weather disasters and the like—but the impetus to a legally enforceable contract or legal liability in tort was closely related to direct human action for most of the history of the law.1 Industrialization introduced the potential for runaway forces that could lead to tort liability, usually through the sudden release of different types of accumulated potential energy (see Losee v. Buchanan, Rylands v. Fletcher), but contracts and organizational action were not likely ever to be significantly driven by nonhuman activity until computing. A developing sense of morality eventually led legal systems to protect some animals to some degree, but animals could not initiate legal action; they could be protected in much the way that private property can be protected, though perhaps for different reasons (Pearson 2011).

That situation has changed dramatically, and the changes continue to accelerate. In the mid-to-late 1900s, it would not have been remarkable for a legal organization to take legally significant internal actions based on automated processes. For example, the bonuses of salespeople could be computed automatically from a complex formula, or salary increases for an employer’s large workforce could be processed without direct human oversight (Bayern 2021). That such processes might not require human activity would have been, at the very least, counterintuitive in the early days of industrialization, but it would have been unremarkable by the turn of the millennium that legally binding conditions in contracts could be triggered or otherwise influenced by software (Bayern 2021). Indeed, by 2000, significant contracts could be made entirely by software—even if software was on both sides of the contract. The Uniform Electronic Transaction Act (UETA), finalized in 1999 and widely adopted by US states, explicitly contemplated and provided for that sort of automation.²

Even under UETA, though, the principals for whom software acted were conventional: they were human beings or traditional types of legal organizations. A further development, a little more than ten years after the Uniform Law Commission promulgated UETA, arose from the advent of cryptocurrency: decentralized “autonomous” organizations could begin to control financial resources without the intermediation of the traditional financial system (Bayern 2014b). Such organizations could, for example, consign to an algorithm the direction of cryptographic assets; an algorithm could direct the transmission of such assets without any further human oversight. At least, they could do so as a factual matter; how law would react to such assets and their transmission remained an open question, but it started to become easier for commentators to see that at least some legally protected property was now subject to what was in a meaningful sense “exclusive” algorithmic control. Legally binding contracts could, moreover, refer to such assets or such algorithms (Bayern 2014a).

A further step that proceeds from these growing technological capabilities was that legally recognized organizations—the only types, or at least the only significant types, of legal persons recognized by law apart from human beings—could begin to defer not only some of their contractual actions but also their internal governance to algorithms. Unrelated directly to technological developments, organizational law in the US had been following a trend that increased the power of organizations’ operating agreements at the expense of ongoing participatory governance by human beings. Corporations, which once required regular meetings of shareholders to elect boards and regular meetings of boards to exercise oversight of the organization, came to permit shareholder agreements and other novel structures that would displace the ongoing power of shareholders and the board (Bayern 2014b, 2021). In other words, a precommitment by participants in an organization could bind the organization and make it very difficult for later participants to change the organization’s structural governance.³ To readers for whom public law is more familiar than corporate law, the effect would be as if constitutions were amended to make it easier for a legislature to bind itself indefinitely and even to prevent future constitutional amendments; corporate law made it possible for an agreement among shareholders at a particular time to govern future affairs of a legal entity despite the potentially divergent wishes of shareholders at a later time (Bayern 2024, pp. 68–76). Eventually, in the late 1900s, LLC law arose and became commonly associated with a principle of “freedom of contract”⁴—that is, freedom of, and potential novelty in, internal organizational-governance mechanisms—so that even the presumption that there might be a board of directors went away, as if it were a product of a different time or at least a different organizational form.⁵

The flexibility of LLC law permitted novel transactional structures. One type of structure arose from the explicit permission in organizational statutes for organizers to adopt immutable operating agreements for LLCs (Bayern 2024, pp. 68–73). Another—just a consequence of the structures enabled or envisioned by particular but widespread types of LLC-enabling statutes—permitted, perhaps surprisingly, the creation of zero-member LLCs.⁶ The result was a legal entity that, though created by humans (or at least by another conventional legal person that was directly controlled by humans), did not require humans for its “perpetual” existence. Instead, the operating agreement—“a contract” in at least the economic sense—controlled the entity. Because of an isomorphism between algorithms and conditions in legal agreements that I pointed out in a line of research started in 2014 (Bayern 2014b, 2019, 2021, 2024), the operating agreement could recognize the verifiable states of software systems. As a result, a software system could control a legal entity without any ongoing internal interference by human beings. Such entities are still, of course, subject to law—traditional human-passed and human-judged rules that govern the external behavior of organizations (e.g., regulating their pollution and holding them liable for torts)—but internally they may be managed entirely by software. If the software is of the decentralized “autonomous” type permitted by blockchains or similar technologies, the factual difficulty of controlling an algorithm couples with the “freedom of contract” of LLCs to produce what is genuinely a legally autonomous entity—that is, an entity free, as a matter of internal organizational governance, from human control.

Such a state of affairs empowers software, enabling it to operate without human beings in exercising the powers of legal entities. For example, software could now enter a contract, serve as a legal agent, or license or transfer legal property. To be sure, the software would not be acting in its own right—the legal system need not recognize any new “rights” of the software or recognize the software as a legal actor—but the action of the software has legal significance because of the way an autonomous entity had been constituted under existing organizational laws. Importantly, though the specific legal developments (in, for example, UETA and the law of LLCs) are local to US law, the factual and legal abilities for an LLC registered in one jurisdiction to operate in another mean that the phenomena I am describing are global in effect.

The final relevant step in the chain of events I am describing is the development of artificially intelligent systems that can respond adaptively to the world. In the 1990s, UETA’s drafters envisioned the possibility of such systems,⁷ as of course had science-fiction authors for decades. But software systems that could pass at least many reasonably constructed versions of the Turing Test were developed in the early 2020s. One relatively small result—though it would not have seemed like a small result even a few decades ago—is that a customer-service “bot” can easily make legally binding decisions on behalf of a conventional corporation or similar organization in responding to customer complaints. But a deeper new result is that a relatively autonomous, adaptively intelligent software system can act in much the role of an independent legal person—as long as it is “wrapped” by a zero-member LLC or any number of other similarly possible creative organizational structures.

That result is possible under existing law without any radical legal reform or even simple statutory reform aimed at artificial intelligence. It emerges from the combination of the capabilities of modern legal organizations and the rapid development of software technology. In other words, through a series of what individually seem like relatively modest changes, it is easy to envision something entirely new under law: an otherwise familiar legal system in which many of the active participants are not human. The ingredients to achieve that result sound almost conventional in retrospect: the “freedom of contract” associated with LLCs, and the novelty of and precommitment in organizational structure associated with that freedom, together enable a new type of legal interaction, and then improvements to the ability of software give new pragmatic power to that type of interaction.

2. Taking Stock of the Current State of Law and Technology

This article reviews the developments toward permitting nonhuman actors to participate in the legal system from the perspective of legal policy. The article aims first, in this section, to evaluate briefly several broad, theoretical approaches to the ability of nonhuman actors to take actions with legal effect; next, in the following section, it suggests several doctrinal categories that may face pressure if nonhuman actors begin to act more widely in the law. The goal and method of the article are to identify potential conceptual problems from a common-law perspective and to lay out potential solutions that common-law courts would have the power to implement, in view of the methods and purposes of the common law (Eisenberg 1988).8 But because the subject matter is to some degree speculative, it is too early to evaluate particular legal cases or judicial opinions.

Two sorts of obvious reactions to this situation, from what can be regarded as opposite perspectives, should be ruled out as appropriate analyses or responses to it. On one hand, it is possible to say that because the legal system has not formally been radically reformed, and because the situation I have described results from private action that is itself relatively unremarkable from the perspective of the law—it is legal and authorized by statute—there is definitionally no problem. This essentially libertarian response is not self-justifying or self-evidently correct, however. For example, a step in a process that externalizes no costs on its own can be aggregated with others in such a way that it causes significant harms that the law ought to recognize (cf. Kahn 1966; Wiseman 2014). As a simple demonstration of that proposition, the emission of the first ton of carbon dioxide from fossil fuels likely causes no problem—not just “no measurable problem” but no problem at all. But in the aggregate, carbon emissions can demand collective action or else risk very significant destruction. In more general terms, traditional legal conceptualism amounts to circular reasoning; it attempts to derive legal conclusions from purported axioms but misses the law’s richness and its ability to respond to change.⁹ To put it differently, an argument that a composition of classically sound conceptual building blocks must be sound is both (1) circular, because of the originally circular justification for the building blocks and (2) incomplete, because the composition itself may raise novel problems.

On the other hand, an automatic or Luddite opposition to new technology is unreasoned and similarly unjustified. For example, a commonplace reaction to the ability for nonhuman systems like artificially intelligent software to take legally binding action often produces responses that sound in values associated with human dignity, as if “dignity” is zero-sum and the replacement of any once-human activity by a machine is a threat (Bayern 2024, pp. 124–33). That view may reflect appropriate fears of abuses of technology, but it is not itself justified; as I put it in a recent paper, it is “equivalent to believing that building a shelter for an animatronic cartoon character, in order to protect it from physical damage, somehow diminishes the value or dignity of shelter for human beings” (Bayern 2026). To be clear, I understand and empathize with those who are unsettled by artificial intelligence and some of its capabilities, but the invitation for AI to work within the system of private law, rather than outside it, does not on its own seem horrific or even inappropriate. Moreover, the harms that AI threatens to cause are not enabled by its interaction with private law; they exist independently from that interaction.

Perhaps too much has depended on the terminology of “legal personhood,” as if “person” conveys a definitional respect beyond the relatively technical capabilities that the private law associates with that term (Bayern 2026). Losing jobs to AI is an important concern; losing friendships to it is arguably even worse, and losing lives to it is definitely worse. But given that it exists, interacting with it through a legal system is not in itself problematic. To make the same point more glibly, if AI destroys the economy by taking everyone’s jobs, or if it comes to threaten the existence of humanity more directly, it has no need to do so by entering into online contracts with people or by opening bank accounts. As I have put it elsewhere:

“My best teacher was a robot” is plausibly a threatening or dispiriting thing for a teacher to hear; “my best friend is a robot” is probably not what one would hope to hear from a close friend and remains an odd thing to hear even from an acquaintance… [But] “I am dispirited because my account at Citibank exists in a database alongside one that identifies a robot” would sound more like a neurosis than a serious concern about law or policy.

(Bayern 2026)

Sometimes technological developments and other sorts of rapid novelty distract from sound legal and policy analyses. I have no grand specific legal prescriptions to offer in this short article, but it would be remarkable if the appropriate response to the increased automation of legal entities and the increased participation of AI in private-law processes like legal contracting ought to be governed by a mindset entirely different from the mindset with which policymakers approach other legal questions.¹⁰ For example, the appropriate response will likely require applying propositions of morality (such as those regarding people’s reasonable reliance on contracts or, for some commentators, the fair distribution of wealth) and policy (such as deterrence of harmful conduct) (Eisenberg 1988) and using some adaptive and probably appropriately undertheorized combination of social morality and welfarism. In saying that, I am simply saying that “law” and its familiar processes can and should continue to govern a state of affairs that involves humans interacting with today’s level of artificially intelligent software.

In discussing, admittedly a bit jocularly, the “decline of anthropocentric law” throughout this article, then, I do not mean to suggest that today’s level of artificial intelligence has any intrinsic worth or moral rights or that the law should not continue to serve human interests. Software may one day develop to the point where it does deserve rights, but such a development remains speculative.¹¹ Human law still has human ends; the point is just that nonhuman systems can become significant active players in that law, making legally binding contracts and other arrangements among themselves or with human beings. It would be extremely premature to prohibit those capabilities, and it would be foolhardy to ignore them.

Legal systems have long adapted to social and technological change. Under a modern view of the common law of contracts and torts, private law excels at such adaptations. It cannot do so when it is bound to conceptualism—when professors envision the purposes of subjects like contract and tort to be the satisfaction of axioms or some abstract, internal conceptual beauty, as if the goal of the law is to please law professors rather than to implement social morality and social policy. Pronouncements at too grand a level of generality—“it is unacceptable for AIs to contract with each other because that violates the proposition that law has human ends”—tend, in the real world, to fail both descriptively and normatively. The important question is how legal processes will and should adapt to address the evolving technological and organizational state of affairs I have described so far.

In case it is helpful to draw the distinction explicitly, my position is that attempting to define or justify new legal capabilities on purely conceptualistic grounds is misguided but that doing so “legalistically”—that is, through the full range of doctrinal, moral, and policy arguments available to modern, dynamic law—is sound. The difference is important but may not immediately be evident to readers from legal traditions that ground “law” or “legal reasoning” in conceptualism. Accordingly, in responding to artificial intelligence, I consider it appropriate to rely on legal reasoning but not on traditional conceptions of legal categories, definitions, or analogies for their own sake.

3. Challenges and Likely Developments in Law

Given the rise of legally significant nonhuman action, several principles throughout private law will likely need to adapt. This section surveys a few of them, with the goal of showing the sort of legal reasoning I have in mind as a response to the increased role of software in the society that private law helps to regulate. My grounding for the discussion is in the common law, but the principles I discuss are likely to apply more broadly.

3.1. Personhood

The first significant change concerns the notion, as discussed briefly earlier, of legal personhood. My work over the last fifteen years has shown that, at least in US organizational law, existing legal entities are sufficient to give a very close analog of private law’s conception of legal personhood to software systems. This sounds threatening to many when they first hear it, but it should not be; legal personhood is not a bestowal of value but rather a technical bookkeeping technique—a simplifying force. That does not imply that everything ought to be a legal person—there may be useful practical constraints for the law to impose on the concept (cf. Bayern 2027)—but personhood is not a matter of desert but of convenience. It would be better to avoid the term person entirely, using instead a term like legally significant system, but that ship has probably sailed. Long ago, theorists floated alternatives to the notion of personhood; Lon Fuller described it as a “question … of terminology merely” (Fuller 1930)¹² and considered whether a term that Frederic Maitland once offered, “right-and-duty-bearing unit,” would be superior (Fuller 1930; Maitland 1911; Bayern 2024). The terminology should not matter; I dwell briefly on the point only to suggest that far too much antipathy toward the notion of legally autonomous organizations appears to be wrapped up in the phrase legal person, as if the legal concept has an almost mystical power.

In any event, the change that I am suggesting that legal systems ought to consider is direct treatment of at least some software systems as legal persons (Bayern 2026). The reason for such treatment is just that it is convenient and tends to simplify legal processes and legal reasoning. The techniques in organizational law that I have developed (Bayern 2021) to let software approximate legal personality are workable, but a cleaner and more honest approach for the law to adopt is just to say that software (e.g., an autonomous organization on a blockchain or an independently funded AI agent running a cloud-software service) can enter contracts and be liable in its own right. That capability is not meant to deny the potential liability of the creators of software or to suggest software should not be regulated; it is just to enable people to do useful things with software as a counterparty, partner, or other legal actor.

Occasionally, in the current generation of AI, commentators aim to deny that intelligent software ought to have this ability to contract. It is hard to square such a position with the policies and practicalities of MBCA, under which electronic contracting is recognized relatively easily and uncontroversially and has caused few problems over a period of decades. In particular, electronic contracting has surmounted two types of conceptual hurdles posed by theorists. One is that novel technological means supposedly demand comprehensive or perfect technological solutions; for example, it was once thought that electronic authentication would demand a sophisticated public-key infrastructure to ensure that those who “sign” electronic contracts can be fully authenticated by technological means (Bayern 2021, pp. 24–26). But the real commercial world has operated fine for decades without comprehensive technological solutions to every problem potentially raised by technology; as businesspeople readily understand, the possibility of some fraud is just one cost among many and does not justify scuttling all advantages of new technology. The second supposed theoretical hurdle is the notion that legal processes once exclusively reserved to humans definitionally (or perhaps morally) require human attributes. As I have suggested so far, this appears to be nothing more than a conclusory assertion; it is not true in fact and would require some further justification to be accepted as a matter of law or policy. The general potential threats posed by AI are not, as far as I can tell, such a justification because they do not give any reason to think the world is safer without the particular use of AI in question.

3.2. Intent

Another significant point of pressure to which the law will likely need to respond is the role of intent in private-law analysis. As software systems take on more significant roles in the world of private law—helping people make contracts and otherwise organize their affairs, potentially incurring tort liability, helping to run organizations, dealing with customers, and so on—rules that depend heavily on the “intent” of counterparties will start to produce bad results (Bayern 2021, pp. 149–61).

For example, as conventionally understood, many torts require intent. If I set up a website that accidentally causes you emotional harm, there is a good chance I avoid liability; your “pure emotional harm” is likely not recoverable under the common law of torts (American Law Institute 2012, § 47). But if I intentionally cause you emotional harm through outrageous behavior, that can amount to the tort of intentional infliction of emotional distress (known in some jurisdictions as outrage) (American Law Institute 2012, § 46). It is proper for the intentional wrongdoer to face special liability; intentional harm is a particular kind of harm that the law can deter and punish, and to the extent we can discern bad mental states, it makes sense to do so. But to rely too heavily on the law’s existing distinctions between intentional and unintentional conduct may cause increasingly significant types of behavior to slip through the cracks. Particularly if businesses use artificially intelligent software to optimize their customer-facing processes, there is a significant chance that the optimizations they adopt will work—without necessarily any human or other discernible intent in specific cases—to the detriment of customers. In other words, setting up a website that optimizes commercial pressure on customers could, at least in theory, mimic the sort of outrageous behavior that constitutes intentional infliction of emotional distress, but it could do so in an intentless way that, for that reason alone, avoids triggering liability for intentional infliction of emotional distress.

My point is not specifically that intentional torts should go away, although there has been ample theoretical critique that many intentional torts are misconceived and could be harmonized into a broader legal framework (Sugarman 2018).¹³ It is that in a world in which technology does more, lawmakers may need to expand liability to cover more real-world behaviors that cause harm. In cases where technology is substituted for employment, there may be a similar need to adapt when tort law imposes vicarious liability on employers for the actions of employees. For example, there is a rarely discussed tort doctrine under which “A common carrier or other public utility is subject to liability to patrons utilizing its facilities for gross insults which reasonably offend them, inflicted by the utility’s servants while otherwise acting within the scope of their employment” (American Law Institute 1965, § 48). Assuming the principle on which that doctrine rests is desirable, it would be relatively easy to adapt the doctrine to include not just employees but technological instrumentalities; for example, a future version of the Restatement should perhaps decouple the notion from employees (“servants”). A similar type of generalization may, more importantly, be needed for the much more general and widely applied doctrine of respondeat superior (American Law Institute 2006, § 2.04).

The same type of reasoning applies to contract law and related subjects. As just one small example, the Restatement (Third) of Restitution and Unjust Enrichment would make the remedy of disgorgement of profits available in some cases where the promisor engages in a “deliberate breach of contract [that] results in profit” (American Law Institute 2011, § 39) The comment to the Restatement explains the requirement that relevant breaches be “deliberate” as being motivated by a desire to exclude cases “in which breach results from the defendant’s inadvertence, negligence, or unsuccessful attempt at performance” and is therefore not “conscious” (American Law Institute 2011, § 39 cmt. f). Even if that choice were the correct one under a conventional state of affairs—a debatable proposition—in a world where more breaches of contract are directed by processes that are not or cannot be “conscious” (in view, at least, of current understandings of consciousness and the current state of AI), the rule may become dramatically underinclusive. To put it differently, simple factual change may put pressure on the law to recognize that a rule that once required “conscious” wrongdoing ought to extend to at least some novel types of unconscious wrongdoing that have similar functional roles to those of conscious wrongdoing except that they are committed by unconscious systems.

A different way for the law governing nonhuman action is to lean toward strict liability over fault-based principles, but that is a more significant change and is beyond my scope in this paper. The distinction between fault-based and no-fault liability has been widely discussed in traditional legal commentary and is likely to retain importance in an era of legally significant action by nonhumans.

4. Some Reflections on, and Speculations About, “Inhuman” Law

My discussion so far has been descriptive and reactive, not strongly evaluative of AI itself and its role in society as mediated by choices of law and policy. To be clear, nothing I say is meant to be “pro-AI” in the sense of being antiregulatory or even of buying into any of the marketing, assumptions, or techniques of the AI industry; I am treating AI as a fact and recognizing that it is functionally, and increasingly formally, a participant in the legal system even without “rights” of its own. I have addressed in separate work how the common law might need to begin to adapt in the event that a type of AI emerges that deserves rights and respect of its own, arguing that such a state of affairs would require legal systems to generate new types of rules to address a population that is “bimodal” in several characteristics (Bayern 2027). But such developments remain speculative and are beyond my scope here; my discussion here has just been about how purely technological “actors” have increasingly taken acts of legal significance. (I intentionally avoid the currently popular terminology of “AI agents” to avoid confusion with the law of agency.)

It is worthwhile to consider several broader possible implications of the increasing attention that law will likely need to pay to nonhuman actors. One such implication derives simply from a kind of shift in perspective. Once we recognize that the legal system—even if it is still serving the same roughly familiar human ends as it has always served—significantly involves nonhuman participants, we should at least consider ways that the law might achieve goals that address or importantly involve nonhuman systems, such as animals or features of nature. It is possible to do this even in a system of law that remains centered on human value, although it is also possible that increased interaction with nonhuman technological artifacts will cause people and eventually policymakers to recognize that they have long had many important interactions with nonhuman systems, that many of those systems have value that does not derive (in at least simple or conventional ways) from human value alone, and that those interactions ought to be regulated differently from the way they have been for the last few hundred years.

It may be worthwhile to consider that the emergence of large language models has shown us that conventional stories we have told ourselves about what makes us “uniquely human” are simplistic and perhaps incorrect. For example, a conventional proposition about humanity was that language was an exclusively human trait; it was one thing that “separated” humans from animals. One way of understanding the scientific contribution of large language models is that they helped to demonstrate that communicating using language was simpler and more attainable through software than science had previously recognized (Wolfram 2023). Language, in other words, is not exclusively human; it is achievable through relatively simple language models already now far behind, as a technological matter, the state of the art. Recognizing that language is not uniquely human may (by removing a supposedly essential feature that distinguishes humans from nonhumans) help us see that we have more in common with animals than we previously recognized—and, indeed, AI may even help us communicate with animals better. So a surprising result of the development of AI could, at least ideally, involve better empathy with animals and increased legal protections for them. Perhaps what is true of language is true of intelligence too; an emerging popular reaction to the state of AI at the time this article was written is that perhaps AI shows that intelligence, in general, arises from evaluative networks more readily than most people thought it would.

To be clear, I am not necessarily suggesting here, as others have,14 that legal personhood should be extended to animals or other nonhuman entities as a mark of respect or concern for their welfare. As I have suggested above, I do not regard legal personhood as a sign of respect or concern; it is largely a bookkeeping mechanism. For functional reasons, perhaps some features of legal personhood, like the ability to own property or to have suits filed in one’s interest, should indeed be extended to nonhuman animals, though many incidents of legal personhood (like those associated with contract law) are unlikely to be relevant. My observation here is only that the underlying respect and concern for animals may follow independently (regardless of formal decisions about personhood) as a result of an increased recognition that the law is engaging a variety of nonhuman participants.

More mundanely, large language models might also help us recognize that law is not the same thing as, and is not fully captured by, language. That language is not uniquely human may, in other words, help us avoid the mistake of linguistic literalism—of thinking that all human affairs can be reduced to language. To put it differently, the emergence of AI can change our imaginative worldview in unexpected ways. Some of those ways are immediately horrific and regrettable; children have already befriended AI systems and harmed themselves as a result of things that the AI systems have said (Kang 2026). But the development of law toward increasing recognition of rights for expanding ranges of systems has generally been a positive force in simple consequentialistic terms, and AI can at least possibly invite legal theorists to imagine new, useful, and even more broadly just directions for the law.