No. As of November 2026, no court or legislature we know of has recognised an AI agent as a legal person, so it cannot be sued or own property in its own name. Two US states, Idaho in 2022 and Utah in 2024, have passed laws against granting AI personhood. When an agent causes harm, the law looks for the people and companies behind it.
A note on the legal and regulatory points here: they summarise public sources as of November 2026 and should not be relied on. Rules for AI agents are still being written, and some of what is described here is proposed or coming soon rather than settled. Check the current position, or ask a qualified professional, before acting on it.
Why can't an AI agent be sued?
Because only legal persons can be sued, and an AI agent is not one. Legal persons are humans and entities the law recognises, such as companies.
The Law Commission of England and Wales' discussion paper on AI and the law, published on 31 July 2025, puts it directly. AI systems have no separate legal personality, so they cannot be sued or prosecuted. The law therefore has to find a natural or legal person to hold accountable.
Some US states have closed the door by statute. Idaho's 2022 law says AI shall not be granted personhood in the state. Utah's law, in force since 1 May 2024, bars government bodies from granting or recognising legal personhood in artificial intelligence, alongside rivers, land and animals (Utah Code 63G-32-102).
Courts have reached the same place from other directions. In December 2023, the UK Supreme Court held that an inventor under the Patents Act 1977 must be a natural person, and that owning an AI machine did not give its owner the patents for what it produced (the UK Supreme Court's press summary). A US federal appeals court has likewise held that AI cannot be a copyright author.
So who gets sued instead?
Whoever stands behind the agent: the user who deployed it, the business that operates it, and in some cases the company that built a defective product. Which one depends on the facts and the country.
| Who | How they might be liable | The catch |
|---|---|---|
| The user or deployer | Their agent acted on their account and their instructions | Often the person who suffered the loss |
| The operator or platform | Negligence, breach of contract, or regulatory duties | Much depends on what their terms of service say |
| The developer or model provider | Product liability for defective software, in some jurisdictions | Narrow scope; see the EU limits below |
| A legal entity the agent runs inside | The entity is the legal person and can be sued | Only as deep as its assets |
The Law Commission paper flags that it may be unclear which parties in an AI supply chain owe a duty of care to a victim. In its example, a healthcare provider using an AI tool is likely to owe its patients one. The position of upstream developers is much less clear. A 2026 academic paper on agent finance makes a similar point: responsibility may be fragmented across users, deployers, model providers and wallet operators, and contracts are one way to allocate it (an academic paper on agent-to-agent finance).
The EU is the furthest along on developers. Its new Product Liability Directive treats software, including AI systems, as a product, and treats AI system providers as manufacturers (the directive's adopted text). It applies to products placed on the market after 9 December 2026. Courts must consider a product's ability to keep learning after release.
There is a limit that matters for money. The directive covers death, personal injury, damage to property and the loss of non-professional data. Pure economic loss is excluded. An agent that simply loses your funds through bad trades does not, on its own, trigger this route.
Can an AI agent own the crypto in its wallet?
It can control a wallet, but control is not ownership. In law, the assets belong to a person: whoever owns the agent, or whoever a court decides holds the property right.
The law on crypto as property is moving, but towards people owning it, not machines. The Property (Digital Assets etc) Act 2025 received Royal Assent on 2 December 2025. It confirms that, under the law of England and Wales, something can be personal property even if it fits neither of the two traditional categories (the Law Commission's note on the Act). Crypto-tokens can fall into that third category. They still need an owner who is a legal person.
On-chain, the agent itself is often property. Under ERC-8004, the draft Ethereum standard for agent identity, each agent's identity is a standard transferable token. Whoever holds that token controls the agent and can hand management to others. So the agent can be bought and sold like any other token, and the legal owner is the holder, not the agent.
For what this means at tax time, see Do I owe taxes on trades my AI agent made?
Can you put a legal wrapper around an agent?
Yes, and that is the closest thing to agent personhood today. You form a company, and the agent helps run it. The company is the legal person; the agent is not.
Wyoming offers one example. A Wyoming DAO LLC, a limited liability company for a decentralised autonomous organisation, can be managed by its members, by algorithm, or both. Its filing must say how far management is algorithmic and give a public identifier for any smart contract used to run it. Like any LLC, it needs a registered agent in the state (the Wyoming Secretary of State's DAO FAQs). Its amended DAO provisions took effect on 9 March 2022.
Whether AI should get its own legal personality is an open debate. The Law Commission paper calls it a "perhaps radical" option and makes no proposals for reform. Arguments in favour include filling gaps in liability. Arguments against include turning AI into a liability shield for developers, and the difficulty of giving an AI assets so it can actually pay. The paper suggests that any such AI might itself need owners, much as companies have shareholders.
What could go wrong?
The real risk is not that an agent sues you. It is that you are harmed and nobody can be made to pay.
Liability gaps. The Law Commission describes a company whose autonomous AI sends false statements to investors, with no employee checking them. Proving the company knew, or was reckless, could be hard. If neither can be shown, there may be no criminal liability at all.
Wrappers as shields. An entity that runs an agent is only worth suing if it has assets. A thinly funded company or DAO can absorb the blame and none of the cost.
Ownership that moves. Because ERC-8004 agent identities are transferable, the party behind an agent can change overnight. Our inference: the operator you trusted may not be the one you would need to pursue.
Financial losses fall outside product law. The EU directive does not cover pure economic loss, which is what most agent failures in crypto cause. Recourse there will usually run through contract, negligence or financial regulation.
Borders. An agent deployed in one country, built in another and trading on a network with no home country raises a basic question: where would you even file? This is unsettled, and it is our judgement that it will stay that way for some time.
The trust takeaway
An agent has no pockets of its own. Before you rely on one, find the person or company whose pockets stand behind it. Check:
- Which legal entity operates this agent, and in which country is it registered?
- What do the terms of service say about losses the agent causes?
- Who holds the agent's identity token and keys today, and can that change without notice?
- Where could you bring a claim, and under which country's law?
- Is there an insurer, a compensation fund or a written refund policy?
If the agent gives investment recommendations, the next question is what duty its operator owes you: see Does an AI financial adviser owe me a duty of care?