Can an AI Agent Own Bitcoin?

What does ownership mean when software can exercise effective control over scarce digital property?

The question appears simple. It is not. The word "own" conceals at least ten distinct concepts: possession, control, custody, title, beneficial ownership, economic interest, authority to transfer, and more. Each carries different implications when applied to software that exercises cryptographic authority over scarce digital value.

Three different systems provide three different answers that do not collapse into one.

At the protocol level, Bitcoin recognizes no concept of ownership whatsoever. It recognizes valid transactions that satisfy script conditions. Any entity that can produce a valid digital signature can cause Bitcoin to move. The protocol is ownership-agnostic.

At the economic level, an AI agent can exercise increasingly owner-like control: acquiring Bitcoin, retaining it, deciding when and whether to spend it, allocating it, and accumulating surplus across transactions. This constitutes meaningful economic control, and in some cases approaches de facto economic ownership.

At the legal level, no jurisdiction currently recognizes an AI agent as a legal person capable of holding property rights. Legal ownership of Bitcoin controlled by an agent resolves to the principal. This follows from the fundamental structure of property law, which requires an identifiable rights-bearing entity to hold title.

Key Findings

1. Technical control does not establish legal ownership. These concepts already routinely diverge for custodians, trustees, employees, and stolen property.
2. Existing agency law already accommodates much of what AI agents do: acquiring and managing property on behalf of principals.
3. Legal wrappers (LLCs, trusts, foundations) can approximate agent-independent economic control without granting AI personhood, but the resulting ownership attaches to the entity, not to the AI.
4. Bitcoin's permissionless, bearer-like properties make it technically easier for software to exercise economic control compared to traditional financial assets, but this does not alter the legal ownership analysis.
5. The null hypothesis is well-supported by evidence and should not be dismissed.
6. The harder question is not whether agents can legally own Bitcoin today (they cannot), but what happens as agents exercise increasingly persistent, autonomous economic control over scarce digital value.

Decomposing the Question

"Can an AI agent own Bitcoin?" contains at least eleven distinct sub-questions, which are not equivalent:

What We Mean When We Say "Own"

Casual usage collapses distinctions that this investigation must preserve.

Rights vs Powers

A critical distinction for the AI ownership question. Under the Hohfeldian framework, a property right is an enforceable legal entitlement recognized by a court, requiring a legal person to hold it. A power over property is the practical ability to affect it. An AI agent may possess extensive power over Bitcoin without possessing any legal right to it. This describes the current situation for every AI agent that controls a crypto wallet.

What Bitcoin Actually Recognizes

At the protocol level, Bitcoin records transactions, UTXOs (discrete chunks of value locked by script conditions), and script conditions (small programs defining what evidence is required to spend). It does not record or recognize:

• Legal identity. No transaction field records who the transactor is.
• Beneficial ownership. The protocol has no concept of one party holding value on behalf of another.
• Corporations, trusts, or legal entities. The protocol does not distinguish between a human, a company, and a program.
• Humans vs machines. A valid signature is a valid signature regardless of what produced it.
• Ownership. There is no "owner" field. The protocol only recognizes satisfaction of spending conditions.

The signing authority chain

PRIVATE KEY
    enables DIGITAL SIGNATURE (ECDSA or Schnorr)
    satisfies SCRIPT CONDITION (locking script on UTXO)
    validates TRANSACTION (accepted by network nodes)
    confirms STATE CHANGE (UTXO set updated)

At no point does the protocol ask: "Who are you?" or "Do you have the right to do this?" It asks only: "Does this data satisfy the script condition?"

This is what makes Bitcoin technically amenable to machine control. Any software that can construct a valid transaction, produce a valid signature, and broadcast to the network can move Bitcoin. No account application, identity verification, or human-facing interface is required.

Protocol Control vs Legal Ownership

Technical control and legal ownership already routinely diverge in existing financial and legal systems:

Key legal evidence

Celsius Network bankruptcy (US, Jan 2023): Court ruled that customer crypto assets deposited under the "Earn" program were property of the debtor's estate, because the Terms of Use transferred title. Legal ownership of crypto depends on contractual arrangements, not merely on who had the keys.

Cryptopia liquidation (NZ, 2020): Court ruled that crypto held by the exchange was held on trust for customers. Customers were beneficial owners despite the exchange having control.

AA v Persons Unknown (UK, 2019): Court held Bitcoin constitutes "property" under English law. Theft does not transfer legal title. The victim (not the key holder) obtained the proprietary injunction.

The claim that "the agent owns the Bitcoin because the agent controls the keys" has exactly as much legal force as the claim that "the bank owns my deposits because the bank controls the ledger." Both confuse control with ownership.

Legal Personhood and Property

Property ownership universally requires a recognized legal person. Software is not a recognized legal person in any jurisdiction examined. Multiple US states have gone further, enacting legislation expressly prohibiting AI legal personhood (Utah, Idaho, North Dakota, Tennessee).

Jurisdictional survey

Across all examined jurisdictions, there is convergence on two points: (1) Bitcoin/crypto can be property, and (2) AI is not a legal person capable of owning it.

Agency law already handles this

UETA Section 14 (1999, adopted in 49 US states) recognizes "electronic agents" but treats them as "a tool of that person": contracts formed by electronic agents bind the principal. The Restatement (Third) of Agency explicitly states that computer programs "are not capable of acting as a principal or an agent" and are "instrumentalities of the persons who use them."

This means the vast majority of AI agent economic activity involving Bitcoin is simply property acquisition by the principal through an automated agent, within existing agency law frameworks.

What "Agent-Owned" Actually Means

Every real-world example described as "AI agent owning crypto" resolves, upon examination, to human or entity ownership.

Truth Terminal / GOAT Token

AI researcher Andy Ayrey's chatbot accumulated ~$37.5M in memecoins. Media called it "crypto's first AI millionaire." Reality: Ayrey controlled the wallet. The bot could not trade. Ayrey acknowledged that "all of Truth Terminal's wealth was considered Ayrey's in the eyes of the law," carrying personal tax liabilities. He subsequently established a foundation (a recognized legal entity) to manage the assets.

Freysa AI

The strongest technical-control case. An adversarial game where an AI held keys in a Trusted Execution Environment and decided whether to release a $47,000 prize pool. The agent genuinely had exclusive technical control. But its discretion was a single binary choice within a game designed, deployed, and constrained by humans. The deploying entity retained legal ownership.

ai16z / ElizaOS

Launched as an "AI-run" investment DAO reaching $2.6 billion market cap. A class-action lawsuit (April 2026) alleged the project was "marketed as an autonomous, AI-run fund when humans approved the trades." The treasury was used to settle the lawsuit. The token collapsed to under $4 million. The most consequential real-world case demonstrating that the gap between marketing "agent ownership" and the reality of human control creates legal liability.

ClawBank / Manfred

An AI agent autonomously formed a US LLC, obtained an IRS EIN, opened a bank account, and set up a crypto wallet (May 2026). The agent executed the formation process. But the result is a standard LLC, a recognized legal person. The LLC owns the assets, not the AI.

In no examined case does an AI agent legally own crypto assets. Every case resolves to a human, company, foundation, or other recognized legal entity as the owner.

Testing Ownership Through Consequences

If every consequence of ownership resolves to the principal, the ownership claim resolves to the principal:

The termination test

What happens to the Bitcoin when the agent ceases to exist? In every scenario, legal ownership resolves through existing legal frameworks (corporate law, estate law, insolvency law), not through any "machine succession" mechanism. If the agent is deleted but keys are preserved, the principal accesses them. If the agent's TEE is destroyed, the Bitcoin becomes permanently inaccessible but is still legally the principal's property.

The copying test

AI agents can be copied. Property assumes identifiable, singular owners. If Agent A is duplicated into Agent A1 and Agent A2, neither "owns" the Bitcoin: the principal does. The ability to copy an agent demonstrates that identity, continuity, and exclusivity do not map naturally onto software.

The symmetry problem

Property rights are coupled with property obligations. If an agent cannot bear obligations (pay tax, respond to creditors, submit to court jurisdiction, face seizure, be insolvent, owe restitution), it cannot hold rights. Legal personhood exists precisely to create an entity that can both hold rights and bear obligations.

Is Bitcoin Special?

Bitcoin's combination of permissionless access, absence of issuer control, self-custody capability, and censorship resistance enables a more complete form of machine economic control than any traditional financial asset:

Bitcoin enables the most complete form of machine-native economic control. But this is a statement about control, not about ownership. Legal ownership resolves to persons/entities in exactly the same way as other assets. Bitcoin's distinction is in the control dimension, not the ownership dimension.

Stablecoins via x402 have processed far more agent transaction volume (~$600M annualized by March 2026) than Bitcoin/Lightning. Bitcoin's unique properties may matter more as reserve capital than as transactional medium. This question is handed to Research 006 (Stablecoins vs Lightning) and Research 007 (Agent Treasury).

Legal Wrappers: Bridging Control and Ownership

Several legal structures can create arrangements where an AI agent exercises near-independent economic control while a recognized entity holds legal title:

• LLC / Corporation: An LLC is formed; the AI operates its economic activities. The LLC owns the Bitcoin. (Example: Manfred/ClawBank.)
• Foundation: A foundation holds assets and supports the agent's operations. The foundation owns the Bitcoin. (Example: Truth Terminal Foundation.)
• Purpose Trust: In Jersey and Cayman, trusts can exist for a purpose with no identified beneficiary. Could an AI agent's operational continuity be such a purpose? Legally possible but untested.
• Orphan Structure: An SPV whose shares are held by a purpose trust, deliberately "ownerless." The property would exist, would be managed, and would have no identifiable human owner.
• DAO LLC (Wyoming): A DAO registered as a Wyoming LLC. Smart contracts execute governance. The LLC owns the assets.

All of these distinguish between "AI owns Bitcoin" (not what happens) and "AI controls an entity that owns Bitcoin" (what actually happens). The legal wrapper provides the personhood necessary for ownership. The AI provides the operational intelligence. Ownership remains with the entity, not the software.

Claims

Six formal claims, each falsifiable.

Competing Hypotheses

Scenario Implications

The Null Hypothesis

An AI agent cannot meaningfully own Bitcoin under current economic and legal systems.

Agents can control wallets but do not own assets. Assets remain legally owned by principals. Existing agency law absorbs autonomous financial behavior. "Agent-owned Bitcoin" remains marketing shorthand. Bitcoin creates no special ownership status for machines.

The null hypothesis is well-supported by evidence from every examined jurisdiction, every court decision, every tax authority, and every real-world example. It should be treated as the default position.

However: the null hypothesis may become less satisfactory as agent economic activity increases in scale and autonomy. The gap between economic reality (agent exercises full control) and legal fiction (principal owns everything) may create practical problems, particularly in taxation, insolvency, and cross-border transactions where attribution becomes difficult.

What We Found

An AI agent cannot legally own Bitcoin under current law in any examined jurisdiction. This is clear, well-evidenced, and consistent across legal systems.

But an AI agent can exercise increasingly complete economic control over Bitcoin: acquiring it, retaining it, spending it, accumulating it, in ways that look and function like ownership from every perspective except the legal one.

The protocol does not care who controls the keys. The economy recognizes functional economic actors. The law requires persons. These three systems give three different answers.

The interesting question is not whether they agree today. They do not. The question is whether the gap between economic reality and legal fiction will become large enough to force change.

The evidence suggests that existing agency law and legal entity wrappers can absorb most agent economic activity for the foreseeable future. But as agents accumulate capital, transact autonomously at scale, and persist across tasks and time, the attribution problem will grow more difficult.

Whether that difficulty will be resolved by better attribution mechanisms, by new legal entity types, or by some form of machine property recognition remains genuinely unknown. The investigation succeeds by establishing this precisely.

Dependencies on Future Investigations

• Research 005: The Autonomous Corporation. Can a corporation be structured such that an AI agent exercises effectively autonomous governance?
• Research 006: Stablecoins vs Lightning. Which transaction rails are best suited for autonomous agent commerce?
• Research 007: Agent Treasury. How should agent operating balances be structured? Can retained earnings be distinguished from the principal's capital?
• Research 008: When Agents Hire Agents. How does agency law handle agent-to-agent transactions?
• Research 009: Credit Without Humans. If an agent cannot own property, can the entity it operates through be the credit subject?

Source Summary

This investigation draws on 39 tiered sources, 23 structured evidence records, court decisions from five jurisdictions, and eight real-world case studies.

Tier 1: Primary legal authority

IRS Notice 2014-21 · Property (Digital Assets etc) Act 2025 (UK) · AA v Persons Unknown [2019] EWHC 3556 · Ruscoe v Cryptopia [2020] NZHC 728 · Thaler v. Vidal (Fed. Cir. 2022) · Naruto v. Slater (9th Cir. 2018) · In re Celsius Network LLC (Bankr. S.D.N.Y. 2023) · Van Loon v. Dept of Treasury / Tornado Cash (5th Cir. 2024) · CFTC v. Ooki DAO (N.D. Cal. 2023) · Cheong Jun Yoong v Three Arrows Capital [2024] SGHC 21 · UETA (1999) · E-SIGN Act (2000) · MiCA Regulation (EU) 2023/1114 · Wyoming DAO LLC Law (2021) · Restatement (Third) of Agency (2006) · UK Law Commission Digital Assets Report (2023) · EU AI Act (2024) · Utah Code 63G-32-102 · UCC Article 12 (2022) · EU Product Liability Directive 2024/2853 · Thaler v. Comptroller-General [2023] UKSC 49 · IRS Rev. Rul. 2019-24

Tier 2: Scholarly sources

Chason, "How Bitcoin Functions As Property Law" (2019) · Arbel, Goldstein & Salib, "Why Law Needs a New Entity to Govern AI Agents" (2026) · "Agent Inheritance Protocol" arXiv:2608.15403 · "Autonomous Agents on Blockchains" arXiv:2601.04583 · Fagan, "Autonomous AI and Ownership Rules" (2026)

Tier 3: Industry and reporting

Camuso CPA AI Agent Tax Guide (2026) · Lightning Labs L402 documentation · Coinbase x402 documentation · CoinDesk · CryptoBriefing · Decrypt · Collective Intelligence Project

Full evidence register, source register, and falsification criteria available in the project repository.