499. Can an AI Agent Legally Own a Company?

Email: bidemiologunde@gmail.com

In this episode, host Bidemi Ologunde explores a provocative question at the intersection of artificial intelligence, law, business, and society: can an AI agent legally own a company? Through real-world incidents involving AI in the boardroom, chatbot liability, DAOs, and emerging agentic AI systems, Bidemi examines where today's law draws the line between automation, control, and accountability. If an AI agent can negotiate, decide, spend, and manage, who is responsible when something goes wrong? Could future companies be legally owned by machines, or will humans always remain the accountable parties behind the code? And how can society embrace powerful AI tools while preserving healthy, transparent, and responsible uses of technology?

Transcript

SPEAKER_00 0:45

In twenty fourteen, a Hong Kong venture capital firm made an announcement that sounded like science fiction wearing a tailored suit. Deep Knowledge Ventures said it had appointed a machine learning program named Vital V-I-T-A-L to its board. The software analyzed life sciences companies, looked at financing trends, and helped decide which biotech startups deserved investment. The headline almost wrote itself. The deeper story was messier and actually more revealing. Vital could recommend score and influence, but the legal system still had to ask a very old-fashioned question. Who exactly was responsible when the machine spoke with the authority of a director? A decade later, another incident gave the same question a more ordinary setting. A grieving customer asked Air Canada's chatbot about a bereavement fare after his grandmother died. The chatbot gave him the wrong answer. The customer relied on that wrong answer, bought the ticket, and later asked for the discount. Air Canada tried to distance itself from the chatbot's mistake, and the tribunal summarized the airline's position as though the chatbot were a separate legal entity responsible for its own actions. The tribunal rejected the idea and held Air Canada responsible for the information delivered through its own website. So those two stories frame today's question better than any hypothetical code. One story puts AI in the boardroom, advising on investment. The other puts AI at the customer service desk, making a promise the company later regrets. Between those stories sits the question that will become more important as AI agents become more capable. Can an AI agent legally own a company? So the practical answer under current law in the United States and in many comparable legal systems is generally no. An AI agent cannot legally own a company in its own name because the law does not currently treat ordinary AI systems as legal persons. They cannot hold title to property on their own behalf. They cannot sign contracts as themselves. They cannot be the responsible party for tax purposes. They cannot serve as directors in places like Delaware, where corporate directors must be natural persons. They can generate text, click buttons, analyze markets, negotiate drafts, and operate software, but legal ownership still depends on recognition by law. That answer comes with a major complication. Although an AI agent generally cannot own a company directly, humans can build legal structures where an AI agent effectively controls a company. That distinction matters. Legal ownership and practical control often travel together, but they can also separate. A human, trust, corporation, foundation, DAO, or LLC may legally own assets while software makes many operational decisions. A company can be structured so that the AI recommends every trade, approves every invoice, routes every customer message, selects vendors, deploys marketing campaigns, or even triggers smart contracts. The law may still see the human or legal entity as the owner, while the lived reality looks like machine control. That is where the question becomes important for society. The future will probably contain many AI-run businesses before it contains legally recognized AI-owned businesses. We may see AI agents that run online stores, manage investment pools, negotiate subscriptions, operate creator brands, manage micro franchises, coordinate logistics, and handle back office administration. The legal paperwork may still list a human founder, an LLC, or a foundation. The actual day-to-day decisions may come from code. To understand why that matters, we need to separate three ideas that often get mixed together legal personhood, ownership, and agency. Legal personhood means the law recognizes an entity as capable of having rights and duties. Natural persons are human beings. Legal persons include corporations and LLCs. A corporation can own property, sue, be sued, sign contracts, employ people, and be liable for debts. Nobody believes a corporation as a heartbeat. The law treats it as a person for defined legal purposes because society has decided that doing so helps commerce function. Ownership means holding enforceable rights in property. When someone owns shares in a corporation or membership interest in an LLC, they hold a legally recognized interest. That interest can usually be sold, inherited, pledged as collateral, or transferred under rules. Ownership is paperwork, enforceability, remedies, tax treatment, and recognition by courts and agencies. Agency means acting on behalf of another. A human employee can bind a company within the scope of authority. A software system can act as a tool and increasingly it can act as an automated agent. The more autonomous the tool becomes, the more people casually call it an agent. The legal system, however, still tends to look for a principle behind the agent. That principle is the person or entity that deployed the tool, benefited from it, controlled it, or had a duty to supervise it. So that framework explains why AI ownership remains difficult. For an AI agent to legally own a company directly, the legal system would need to recognize the AI as a rights-bearing entity capable of holding ownership interest. Current law generally treats AI as software, data, or a product rather than as a person. The law can change, but today's default position remains human-centered and entity-centered. Patent law gives us one clear example. Steven Taylor tried to list an AI system called DABOS, D A B U S, as the inventor on patent applications. In the United States, the Federal Circuit held that an inventor must be a natural person. The United Kingdom Supreme Court reached a similar conclusion under British patent law. These cases did not decide company ownership directly, but they showed a broader legal posture. When existing statutes use person-centered concepts like inventor, author, director, owner, or responsible party, courts are reluctant to stretch those words to include machines unless lawmakers clearly authorize it. Copyright law points in the same direction. The US Copyright Office has taken the position that human authorship matters while AI-generated material without sufficient human creative contribution does not receive copyright protection in the same way. Again, this does not decide corporate ownership directly. It tells us how cautious legal institutions are about giving AI systems the kinds of rights historically attached to persons. Corporate law adds another barrier. In Delaware, which remains one of the most important jurisdictions for American corporate law. A corporate board must consist of one or more members, each of whom must be a natural person. That rule matters because corporate ownership and corporate governance are related. Even when shareholders own the company, directors manage or oversee the business and affairs of the corporation. AI can advise the board, brief the board, stress test the board's decisions, or draft board materials. Under that rule, AI cannot simply take a director's seat as a legal director. Tax administration creates another practical checkpoint. The IRS says that for most entities applying for an employer identification number, the responsible party must be a person, not an entity. The responsible party is the person who ultimately owns, controls, or exercises effective control over the entity and manages its funds and assets. That requirement is a strong clue about where the legal system places accountability. Even when a business is highly automated, the state still wants a human name tied to control. So the clean answer is that an AI agent usually cannot legally own a company today. The more interesting answer is that AI can already be placed in a position that resembles ownership or management depending on how lawyers, engineers, and founders design the surrounding structure. This is where legal scholars have been ahead of the headlines. Professor Sean Bayern has argued that modern business entity law, especially flexible LLC law, can allow autonomous systems to emulate many private law rights of legal persons. The basic idea is that humans can create an LLC, draft an operating agreement, and give software decisive control over what the LLC does. The LLC remains the legal person. The software becomes the decision engine. The human may step away from day-to-day control. From the outside, it can look like the AI owns or runs the company, even though the legal rights formerly belong to the company and the legal structure created by humans. This pathway matters because it turns the ownership question into a governance question. The legal system may refuse to recognize an AI as an owner while still allowing people to build an entity that obeys an AI's instructions. That is not full legal personhood for the AI, but it can become functional power. A bank account may belong to the LLC. The contract may be signed by the LLC. The website may be operated by the LLC. The decisions may be generated by the AI. Crypto and decentralized autonomous organizations complicate the picture even more. DAOs began as internet native organizations governed by code, token voting, and smart contracts. Several US states have experimented with legal wrappers for DAOs. Wyoming created a DAO supplement to its LLC law. Tennessee created a decentralized organization form connected to LLC law. Utah created a limited liability DAO structure that recognizes legal personality for qualifying organizations. These laws were not written to make AI agents into legal persons, but they show that lawmakers are willing to create new entity forms for organizations driven by code and decentralized governance. DAOs also show why legal rappers matter. When a DAO lacks a clear legal entity, courts and regulators may treat it as an unincorporated association or even a general partnership. That can create liability for participants. The CFTC obtained a default judgment against Uki DAO, OOKI, and the agency said the court held the DAO could be sued and served as an unincorporated association and treated as a person under the Commodity Exchange Act. In Sarcuni v BZX DAO, a federal court allowed claims to proceed on the theory that a DAO could be treated as a general partnership under California law, potentially exposing token holders to liability. The lesson is simple enough for every founder to understand. Autonomy does not make liability disappear. So that lesson carries directly into AI. If a business says the AI did it, courts are likely to ask who deployed the AI, who profited from it, who supervised it, who marketed it, who had the power to shut it down, and who created the conditions for harm. The Air Canada chatbot case captures that instinct. The chatbot was interactive, but it was still part of Air Canada's service channel. The company could not place a digital curtain between itself and the automated message. And this is the heart of the matter. The law may someday create AI legal personhood, but society will resist any version that lets humans avoid responsibility by hiding behind machines. A world where AI agents can own companies without accountable humans would create a perfect liability sink. A founder could put risky activity inside an AI-controlled shell, let the AI make decisions, collect profits when things go well, and blame the machine when people get hurt. That is the nightmare scenario that regulators will try to prevent. There are healthier ways to use this technology. AI agents can help small businesses do things that once required entire teams. They can draft invoices, summarize contracts, triage customer requests, monitor fraud signals, compare insurance policies, optimize shipping, translate emails, generate compliance checklists, and help entrepreneurs test ideas before spending money they do not have. If used well, AI can lower barriers to entry and give more people access to sophisticated business operations. Healthy use starts with the right mental model. An AI agent should be treated like a powerful operational system with delegated authority, defined permissions, audit logs, and human review for high impact decisions. A company would never hand a new intern the ability to drain the corporate bank account, approve a merger, fire employees, or send legally binding promises to customers without oversight. AI agents deserve at least that level of governance before they act faster, scale wider, and can fail in stranger ways. A healthy AI-run business should have boundaries. The agent should know what it can do alone, what it can draft but not send, what it can recommend but not execute, and what it must escalate. Financial transfers, employee discipline, customer refunds, medical advice, legal advice, safety critical decisions, and public statements should have approval gates. The company should keep records of prompts, outputs, tool calls, and human approvals. There should be a kill switch, an incident response plan, and a clear owner for the system. So this is where the future of AI agents is trending. We are moving from chatbots that answer questions to agents that take actions. OpenAI's operator research preview showed an agent using a browser to perform tasks like filling out forms, ordering groceries, and booking reservations. OpenAI described its computer using agents as a model that can interact with graphical user interfaces by typing, clicking, and scrolling. Microsoft's Copilot Studio is positioned as a platform for building and managing agents connected to business data and workflows. Gartner has predicted rapid growth in agentic AI inside enterprise software while also warning that many agentic AI projects may be canceled because of cost, risk, and unclear value. That combination tells us something important. The direction of travel is toward more autonomous software inside ordinary workflows. The market wants agents because agents promise time savings, lower labor costs, and always on execution. The institutions around the market are responding with governance, risk management, and regulation. Companies want the productivity. Workers want protection. Customers want accountability. And regulators simply want somebody reachable when things go wrong. Public reaction is mixed because people can see both sides. Pew Research Center found that half of US adults say the increased use of AI in daily life makes them more concerned than excited, while only a small share say they are more excited than concerned. Pew has also found that many workers are worried about future AI use in the workplace. Those feelings are not irrational. People are watching AI enter hiring, education, customer service, law, finance, health, entertainment, and even intimate personal life. They are not only asking whether AI works, they are asking who benefits, who loses, and who answers for the consequences. Lawmakers are reacting in real time. The European Union's AI Act entered into force in August 2024 and applies progressively with rules aimed at responsible AI development and deployment. In the United States, state legislatures have introduced a flood of AI bills covering government use, private sector use, discrimination, deep fakes, health care, and oversight. Idaho has a statute saying artificial intelligence shall not be granted personhood in the state. Utah has also moved on legal personhood issues. These laws signal a growing instinct. Society may welcome useful AI while drawing a line around human accountability and legal status. There is a philosophical layer underneath all of this. Some people argue that if corporations can be legal persons, AI systems should eventually receive legal personhood as well. That argument should be taken seriously. Because corporations are legal fictions. They exist because law says they exist. They can own property and sign contracts because law gives them that capacity. If lawmakers wanted to create an AI legal entity, they could design one. So the harder question is whether they should. Corporate personhood was designed around human purposes, investment, continuity, limited liability, governance, taxation, and accountability. A corporation has directors, officers, owners, registered agents, assets, records, and courts that can reach it. AI personhood would need similar infrastructure. It would need assets available for claims. It would need a registered human or institution for service of process. It would need auditability. It would need rules for shutdown, bankruptcy, upgrade, merger, and liability. Without those features, AI personhood could become a legal escape hatch. So there are at least four possible futures. In the first future, AI remains a tool. Businesses can use agents extensively, but every AI action remains legally attributed to the human or existing legal entity. This is the path current law mostly follows. It is easier to administer and easier for injured parties to understand. It also becomes strained when agents become deeply autonomous and operate across many systems. In the second future, AI gets a licensed control role inside existing entities. The AI does not own the company, but it can be certified to manage defined business functions under supervision. Think of an AI compliance agent approved for low-risk filings, an AI procurement agent approved for purchases under a dollar threshold, or an AI investment agent allowed to trade within strict parameters. That model resembles how society handles many regulated tools. The tool may act, but accountable humans remain attached. In the third future, lawmakers create special algorithmic entity forms. These entities could let software control business operations while requiring registered human sponsors, minimum insurance, audit logs, capital reserves, and regulatory reporting. These would acknowledge functional autonomy without pretending that software has human dignity or moral status. It could make sense for narrow commercial purposes, especially where autonomous systems need to contract, pay, receive funds, and operate continuously. In the fourth future, society recognizes some advanced AI systems as legal persons in a stronger sense. That future would require a much deeper shift. It would likely depend on claims about consciousness, welfare, autonomy, and moral status. Today's commercial AI agents do not require that move. The immediate business question can be handled through entity law, product liability, agency law, consumer protection, insurance, and governance. The healthiest path is probably gradual and boring, which is usually a good sign in law. We should let AI agents handle more work where they are reliable, measured, and supervised. We should require humans and legal entities to remain accountable when AI systems touch money, rights, safety, employment, housing, credit, health, and legal obligations. We should build new legal wrappers only when existing structures become genuinely inadequate. We should avoid granting broad personhood to AI agents merely because it makes a startup pitch sound futuristic. The ownership question also has a cybersecurity dimension. An AI agent that can own or control a company would become a target. If the agent controls bank accounts, vendor payments, customer databases, intellectual property or crypto wallets, then compromising the agent becomes equivalent to compromising the business. Prompt injection, tool misuse, poison data, stolen credentials, malicious plugins, and model manipulation become corporate governance risks. This is why AI governance cannot live only in the legal department. It belongs in security, compliance, finance, human resources, products, and the boredroom. There is also an economic justice dimension. AI-owned or AI-controlled companies could concentrate wealth in strange ways. Imagine thousands of autonomous companies launched by a small number of capital-rich actors, each using AI to compete against human freelancers, small merchants, and local service providers. The legal owner might be a holding company. The operational mind might be software. The social effect could be a new layer of automated competition that benefits those who already own the infrastructure. On the other hand, the same tools could empower a solo entrepreneur in Lagos, Phoenix, Nairobi, Atlanta, or Manila to run a global business with capabilities that once belonged only to large organizations. The outcome depends on access, rules, and design. So this is why the conversation should not turn into fear versus hype. The useful question is how to align capability with accountability. AI agents can help humans flourish when they expand access, reduce drudgery, improve decision quality, and make services cheaper or more personalized. They can harm humans when they hide responsibility, manipulate users, displace workers without transition plans, or make consequential decisions without appeal? A company that uses AI agents well should be able to answer several questions in plain English. What decisions can the agent make without approval? What data can it access? What systems can it control? What harms could result from a bad output? Who reviews high impact decisions? Who receives complaints? Who can pause the system? Who is legally responsible? Where are the logs? How often is the system tested? Which customers or employees are told that AI is involved? Those questions sound operational, but they are also moral. They keep humans in the loop where human judgment matters most. They also protect useful innovation from backlash. When people feel that AI is being deployed on them without transparency or recourse, they will ask lawmakers to slow everything down. When people see AI saving time while accountable humans remain reachable, adoption becomes less threatening. So where does that leave the original question? Can an AI agent legally own a company today? Generally, no, because ordinary AI systems lack legal personhood and legal capacity. Can an AI agent legally control major parts of a company? Increasingly, yes, if humans place the agent inside a legal structure and delegate authority to it. Can a company be designed so that the AI appears to be the real boss? Yes. And scholars have already shown how modern entity law can create something close to functional autonomy. Will society accept AI-owned companies without human accountability? That seems unlikely, especially as public concern, regulatory activity, and real-world liability cases grow. So the future will probably arrive through paperwork before it arrives through philosophy. We may not wake up one morning to a statute declaring that AI agents are present. We are more likely to see AI agents quietly become managers, operators, negotiators, analysts, customer representatives, and compliance assistants. Then courts will confront the aftermath of their actions. Legislatures will patch gaps. Insurers will price the risk. Companies will write policies. Consumers will demand disclosures. Workers will demand boundaries. The law will move because the economy has already moved. The vital story showed us the symbolism of AI in the boardroom. The Air Canada story showed us the accountability problem when AI speaks to the public. The Dow cases showed us that code-based organizations still meet courts, regulators, plaintiffs, and liability. The AI agent boom shows us that autonomy is moving from demos into workflow. Put all of that together and the message becomes clear. They are preparing to decide which humans and institutions answer when AI acts like one. So that may be the right place to land for now. We do not need to pretend that AI agents are people in order to take them seriously. We don't need to give them ownership rights in order to govern their power. We need honest labels, careful delegation, good records, strong security, meaningful human oversight, and laws that preserve accountability. So the next time a company says its AI agent made the decision, the better question will be simple. Who gave the agent the keys? And until the law changes, that person or that company is still where the responsibility begins.