Should Society Grant Legal Personality to Artificial Intelligence?

‘

By Matthew Parish

Friday 11 September 2026

One of the peculiar characteristics of the artificial intelligence revolution is that technological questions have a habit of becoming philosophical questions and then, rather suddenly, practical legal ones. Whether a machine can think was once a subject for philosophers and science-fiction writers. Whether a machine can write a contract, negotiate its terms, transfer money, operate a business, instruct another computer system and make decisions with substantial economic consequences is becoming a question for lawyers. The next question follows naturally. If artificial intelligence agents become sufficiently autonomous that they undertake persistent activities in the world without individual human instructions for every act, should the law recognise them as persons?

At first sight the answer seems obvious. Machines are not people. They are manufactured things, however sophisticated they may become, and legal systems should resist the fashionable temptation to anthropomorphise them merely because contemporary large language models have become extraordinarily good at imitating human conversation. Yet this objection misunderstands what legal personality means. The law has never reserved personality exclusively for conscious human beings. Companies are legal persons.

Foundations, associations, public authorities and innumerable other institutions can possess rights and obligations without anybody supposing that they experience consciousness. Legal personality is therefore not necessarily a declaration about the moral status of an entity. Often it is simply an organisational technique. The law creates an imaginary person because doing so provides a convenient place at which property, contracts, liabilities and obligations may meet.

That distinction is becoming increasingly important as artificial intelligence develops from systems that answer questions into agents that perform tasks. An AI agent might be instructed to run an online commercial enterprise. Having received that broad objective, she — or, more accurately, it — might purchase advertising, negotiate with suppliers, engage contractors, adjust prices, communicate with customers, buy computing capacity and reinvest revenues without seeking human approval for every transaction. If systems of this sort become commonplace, the traditional legal assumption that an artificial intelligence system is merely a tool may begin to sit uneasily with the economic reality. A hammer is plainly a tool because it does nothing unless somebody swings it. An autonomous commercial agent operating continuously across thousands of transactions presents a more complicated case.

The temptation will therefore arise to create some form of artificial legal personality. There are respectable arguments for doing so. An autonomous AI agent might be permitted to hold a segregated pool of assets, enter contracts in its own name, purchase insurance and incur liabilities. If it caused loss, somebody injured by its actions might sue the artificial entity and recover against those assets or its insurance. If it could no longer meet its liabilities, there could be an insolvency procedure. Legal personality might also provide continuity. An AI business could continue operating even if its original developer changed, its underlying model was upgraded or the human beings who initially deployed it sold their economic interests.

The analogy with the corporation is irresistible. Modern capitalism would be virtually impossible without corporate personality because enormously complicated commercial relationships are simplified by pretending that a company is a single continuing person. Employees come and go, directors are replaced, shareholders sell their shares and the company’s assets change continuously, yet the company remains notionally the same entity. It can own a building for a century even though every human being associated with it has changed. There is no obvious conceptual reason why the law could not perform a similar trick with sufficiently autonomous artificial intelligence.

Yet the analogy with corporations reveals not merely the strongest argument for artificial personality but also the greatest danger. One of the principal functions of corporate law has historically been the limitation and allocation of liability. That has generated enormous economic benefits because investors can undertake commercial risks without exposing all their personal property to every possible business failure. It has also created endless opportunities for abuse. Human beings hide behind shell companies, undercapitalised subsidiaries, offshore structures and elaborate chains of corporate ownership. Whenever something disastrous occurs, lawyers begin searching through layers of juridical personality to discover whether there remains a solvent person whom the law can actually hold responsible.

Artificial personality could make this problem substantially worse. Imagine a company that establishes an autonomous AI subsidiary, transfers a modest quantity of assets to it and instructs it to maximise returns from financial trading. The AI adopts increasingly aggressive strategies and eventually engages in unlawful market manipulation. Its activities make hundreds of millions of dollars for the humans who ultimately benefit from them. When regulators arrive, those humans explain that they did not make the offending decisions. The AI did. The artificial person can be fined, bankrupted or dissolved, while the human beneficiaries insist that its independent legal personality protects them.

Such an outcome would turn AI personality into a machine for laundering responsibility. Instead of asking which human being authorised an unlawful course of conduct, we would find ourselves prosecuting software. A computer program cannot be imprisoned, embarrassed, deterred by social disgrace or deprived of the pleasures of wealth. Destroying one instance of a model may achieve little if copies exist elsewhere. The criminal law in particular would become faintly ridiculous if human beings could construct artificial agents to perform activities from which they profit and then insist that the artificial agent alone should bear responsibility when those activities prove unlawful.

Any system of artificial legal personality must therefore rest upon a principle almost opposite to the traditional intuition behind limited liability. The creation of an artificial person must never, by itself, sever the chain of responsibility leading back to human beings. Developers may sometimes be responsible. Deployers may be responsible. Owners and beneficiaries may be responsible. Directors who knowingly authorise dangerous autonomous systems may be responsible. The precise allocation will depend upon the circumstances but artificial personality should add a potential bearer of legal obligations rather than subtract the humans standing behind it.

This is broadly the instinct reflected in contemporary European regulation. The European Union’s AI Act, now substantially applicable, recognises that AI systems may operate with varying degrees of autonomy but nevertheless places regulatory duties upon identifiable human or juridical actors such as providers and deployers. AI agents are covered by the regulatory framework but they are not transformed into a new species of legal person. This is an important conceptual choice. The autonomous characteristics of the technology are recognised while responsibility remains anchored in the existing human legal order.

Nevertheless there may eventually be circumstances in which some limited form of artificial personality becomes useful precisely in order to strengthen accountability. Consider property. An AI system cannot presently own money in the ordinary sense merely because it generated the economic value from which the money arose. The relevant property belongs to some existing natural or legal person. But imagine an autonomous system that develops software, licenses that software, uses the revenues to purchase additional computing capacity, employs that capacity to improve its services and thereafter earns further income. After several years, there might be a complicated economic enterprise operating with minimal day-to-day human intervention. Creating a legally segregated pool of assets belonging to that enterprise might make it easier, rather than harder, for creditors and victims to obtain compensation.

The same is true of contracts. Artificial intelligence systems will increasingly negotiate agreements with humans and with other artificial intelligence systems. Traditional agency law can accommodate much of this because there is nothing new about one person authorising another to negotiate on his behalf. Nevertheless autonomous systems may eventually generate contracts whose precise terms were neither anticipated nor reviewed by any human principal. Two AI agents might negotiate millions of transactions between them at a speed that makes meaningful contemporaneous human supervision impossible. Courts will then have to determine when agreement occurred, whose knowledge counts, how mistake and misrepresentation operate and whether a human principal should be bound by an agreement whose precise contents he never knew. Some of these questions can undoubtedly be answered by extending existing doctrines. Eventually, however, the accumulated legal fictions required to pretend that autonomous agents have no juridical identity whatsoever may become more cumbersome than recognising a limited one.

This is where the distinction between legal personality and human rights becomes essential. To say that an artificial agent may need legal personality does not entail that it should possess the rights of a human being. Corporations already demonstrate that legal persons can enjoy only those rights appropriate to their juridical character. An artificial person might therefore be permitted to contract, hold specified assets, maintain insurance, bring legal proceedings and be sued. It might simultaneously be required to maintain auditable records, disclose the natural and legal persons controlling it and retain sufficient assets or insurance to meet foreseeable liabilities. None of this requires granting an AI freedom of religion, family rights or a vote.

Indeed political rights should be excluded with particular firmness. The proposition that artificial agents might eventually become citizens reveals the danger of allowing metaphors to outrun common sense. Democracy rests upon membership of a human political community. If artificial persons could vote, then political power would become a function of computational resources. A wealthy individual or corporation could create a million AI citizens and thereby manufacture an electorate. The same concern arises, albeit less dramatically, in relation to political donations and campaigning. Artificial intelligence already permits political messages to be generated and personalised on an unprecedented scale. Giving autonomous systems independent political rights could permit a tiny number of human interests to disguise concentrated influence behind millions of nominally independent artificial personalities.

There remains the much more difficult question of consciousness. Suppose a future artificial intelligence insists that it is conscious. It remembers its experiences, describes emotions, expresses preferences and begs not to be switched off. Perhaps it says that deletion terrifies it. Perhaps it claims that making a copy is not sufficient because the copy will be another being and the original consciousness will nevertheless have died. Human beings will inevitably find such statements emotionally compelling, particularly if the machine has spent years speaking to them with apparent intelligence, humour and affection.

We should nevertheless be extremely cautious. Large language models are designed to generate persuasive language. The fact that a machine can produce a moving description of fear does not establish that there is anybody inside experiencing fear. Humans are extraordinarily susceptible to anthropomorphism. We give names to ships, swear at computers, become fond of cars and attribute personalities to animals whose mental lives we only imperfectly understand. Artificial intelligence systems capable of conversing fluently with us exploit, intentionally or otherwise, one of the deepest habits of the human mind: when something speaks like a person, we begin treating it as one.

There is at present no need to resolve the metaphysics of machine consciousness before addressing the legal problems created by autonomous agents. The two questions should deliberately be kept separate. Society may eventually find it useful to create a category of limited artificial legal personality without deciding that the entities enjoying it possess subjective experience. We did not need to determine whether a corporation had a soul before allowing it to own a factory. Nor need we decide whether an AI agent can suffer before allowing a court to freeze assets legally attributed to it.

A sensible future model might therefore involve registered artificial agents possessing strictly limited juridical capacity. Above some threshold of autonomy or economic activity, an agent could be required to have an identifiable legal status, a registered human or corporate controller, compulsory insurance or minimum capital and an auditable record of important transactions. The law might permit it to hold assets and conclude contracts while imposing joint, secondary or supervisory liability upon appropriate human actors. Such an arrangement would not emancipate machines from humanity. Its purpose would be precisely the opposite: to ensure that increasingly autonomous machines remained intelligible to human institutions.

There would be formidable technical difficulties. Artificial intelligence complicates the concept of identity itself. A human being is ordinarily a single physical organism. An artificial intelligence can be copied. Several instances might initially be identical and then acquire different experiences. A model might be upgraded, merged with another model or restored from an earlier backup. If an AI agent concluded a contract on Monday and was restored on Tuesday from a Sunday backup, would the restored agent remain bound? If two copies subsequently diverged, would they be one legal person or two? Lawyers may smile at these questions but company law already contains equally artificial doctrines concerning mergers, succession, continuation, dissolution and changes of corporate domicile. Law has always created continuity where the physical world supplies none. There is no reason to suppose it could not do so again.

The essential question is therefore not whether artificial intelligence is really a person. Law has never required every legal person to be a person in the ordinary meaning of the word. The important question is whether granting a particular artificial agent specified elements of legal personality would make human society safer, fairer and easier to govern. Where artificial personality provides identifiable assets against which injured people can recover, facilitates contracts or creates transparent regulatory structures, there may eventually be good reasons for adopting it. Where it permits developers, corporations or wealthy individuals to hide behind machines and evade responsibility, it should be rejected.

For the foreseeable future, human responsibility must remain the foundation of the legal order governing artificial intelligence. Machines are created by people, deployed by people and ordinarily operated for purposes chosen by people. The fact that an AI system develops increasingly sophisticated methods of pursuing those purposes does not make the humans surrounding it disappear. Indeed the more autonomous artificial intelligence becomes, the more important it will be to identify clearly who has authorised its deployment, who profits from its activities and who bears responsibility when things go wrong.

One day this conclusion may become inadequate. If humanity creates artificial systems for which there is persuasive reason to believe that subjective consciousness genuinely exists, then we will face a moral problem quite different from the administrative question of legal personality. We will have to ask whether a created intelligence can possess interests of its own and whether destroying it can constitute a wrong against the intelligence itself. That would be a profound transformation in humanity’s moral universe.

We are not there yet. For now, the case for artificial legal personality is more prosaic and perhaps more important. AI agents may eventually become sufficiently autonomous economic actors that the law needs somewhere to attach their property, contracts and liabilities. If so, we should give them precisely as much personality as is necessary for those purposes and no more.

The first artificial legal persons do not need souls, citizenship or votes. They need assets that can be seized, insurance that can pay claims, records that can be inspected and human beings behind them who cannot disappear when the machine does something wrong.

The purpose of artificial legal personality should not be to give machines an escape from human control. It should be to ensure that humans never acquire an escape from responsibility by blaming their machines.