No. An AI agent can control money, but it cannot legally own it. It can hold the private key to a crypto wallet and move funds, yet AI has no legal personality: the Law Commission for England and Wales said so in July 2025, and since May 2024 Utah law has barred its government from recognising AI as a legal person. A person or company stays the owner, and stays answerable.
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.
What is the difference between controlling money and owning it?
Control is who can move the money; ownership is who has the legal right to it. In everyday banking the two usually line up. In crypto, an agent can split them apart.
Control in crypto comes from a private key, the secret number that signs transactions. The SEC’s investor bulletin on custody explains that the private key is what authorises transactions for the assets in a wallet. A blockchain checks the signature, not the signer’s identity.
So if software holds the key, software has control. But a blockchain has no concept of legal title. The question of who owns the funds is answered off-chain, by law and contracts.
| Control | Ownership | Responsibility | |
|---|---|---|---|
| What it means | Who can move the funds | Who has the legal right to them | Who answers when it goes wrong |
| Where it is decided | On-chain, by whoever holds the key | Off-chain, by law and contracts | Off-chain, by law, contracts and terms of service |
| Can an AI agent have it? | Yes | No | No |
Why can’t an AI agent be the legal owner?
Because owning property requires being a legal person, and no AI system is one. Legal personality is the status that lets someone hold rights and duties: to own things, sign contracts, sue and be sued. Humans have it, and so do companies.
In its July 2025 discussion paper, AI and the Law, the Law Commission for England and Wales noted that autonomous and adaptive AI systems do not currently have separate legal personality. It listed the ability to own property among the features legal personality commonly brings. It raised granting AI some form of legal personality as a possible future option, while saying current systems may not be advanced enough to warrant it.
Some lawmakers have gone further and closed the door. A Utah statute effective 1 May 2024 bars government bodies in the state from granting or recognising legal personhood in artificial intelligence, alongside things like bodies of water and non-human animals.
Companies are the closest comparison, and they show the gap. A company is a legal person, but people direct it and answer for it. An agent has no such legal wrapper unless someone builds one around it, and that brings the question straight back to a human or a company.
The practical upshot is simple. Whatever an agent “has”, someone else legally owns.
So who owns the money an agent controls?
The owner is whoever stands behind the agent: the user who funded it, the developer who deployed it or the company that runs it. Which one depends on how the setup is built and what the contracts say.
Three common arrangements:
- The agent uses your wallet. You grant it permission to sign some transactions. The funds are plainly yours, and so are its mistakes.
- The agent has its own wallet, funded by you. The key sits with the agent’s software, but the money came from you, and the agent cannot be its legal owner. In practice, the agent can still spend it.
- The agent belongs to a company. A business deploys the agent and funds its wallet. The company owns the funds and carries the risk.
On-chain identity standards make the ownership chain visible. A draft Ethereum standard for agent identity, ERC-8004, registers each agent as a transferable token. The token’s holder is the agent’s owner and can transfer it or delegate management. The agent’s payment wallet starts as the owner’s own address, and the standard clears it whenever ownership changes hands, so the new owner must set it again.
In other words, even the standards built for agents assume a human or company at the top. That is our reading of the design, and we think it is the right one.
Who should hold the keys?
The keys can sit with you, with the agent’s software or with a custodian, a firm that safeguards keys for others. Each choice shifts the risk.
Self-custody means you hold the key. The SEC bulletin notes that you then carry sole responsibility for its security, and that a lost key means permanently losing access to the assets.
Agent-held keys mean the software holds a key and signs on its own. That is what makes autonomy possible, and also what makes a compromised agent dangerous.
Third-party custody means a custodian manages the keys. The SEC bulletin flags questions to ask, including whether the custodian uses customer assets for its own lending, called rehypothecation, and whether it pools customer assets together, called commingling.
A common middle path is a policy-controlled wallet. The agent can propose or sign transactions, but only within rules a human sets, such as spending caps and approved recipients.
None of these choices changes who owns the funds. They change who can move them, and how far a mistake can spread before a human notices.
What could go wrong?
The danger is the gap between control and accountability. When software controls money that it cannot legally own, it can be unclear who answers for a loss.
The Law Commission describes possible “liability gaps”: cases where no person or company is liable for harm caused by an AI system. It stresses that such gaps are a possibility rather than a certainty, and asks whether any are serious enough to need reform. For anyone handing an agent a wallet, “possible” is enough to plan around.
Specific risks to watch:
- Unclear ownership. If an agent’s wallet was set up by a developer but funded by you, who has the claim if funds go missing?
- Key compromise. Anyone who takes over the agent’s software takes over its key.
- Final transactions. A blockchain will not reverse a transfer just because the agent had no authority to make it.
- Weak paperwork. Terms of service may say little about losses caused by an agent acting on its own.
Crypto users are split on this. An April 2025 survey of 2,632 crypto users by a major crypto data platform found 34.5% would trust AI with wallet access, while 37.5% would not. The other 27.9% were neutral, which in our reading means many users have not yet decided where the line sits.
The trust takeaway
An agent can hold the keys, but it cannot hold the responsibility. Before you let an agent control any money, find out who owns the funds, who holds the keys and who answers if something goes wrong.
- Owner: Whose name is legally attached to the funds the agent controls?
- Keys: Who holds the private key, and could the agent move funds without your approval?
- Limits: Are spending caps and approved recipients enforced by the wallet, not just promised?
- Custodian: If a third party holds the keys, does it lend out or pool customer assets?
- Recourse: Do the terms say who is liable for an agent’s mistakes?
For organisations, this is a custody question first. Panthea, TrustSwap’s institutional custody offering, safeguards digital assets for institutions. Where this is going: custody policies under which an agent can act within limits set and owned by a human, so control can be delegated without giving up accountability.