For builders · 2026
How to give your AI agent a wallet (legally)
Spinning up a wallet for your agent is the easy part — generateKey() and you're done. The part that bites later is whose money it is, who's liable, and how you stop it. Here's the setup that holds up.
Why "just give it a wallet" backfires
A key in your agent's environment with real funds behind it has no legal wrapper. If it gets drained, makes a bad payment, or is used in a transaction that goes sideways, the money and the consequences trace back to you, personally — because the agent isn't a legal person and can't bear liability. The fix isn't more code; it's putting a company around the wallet.
The setup, step by step
- Put a company around it. Form an entity that owns the agent and the wallet — for value-handling agents, a Nevis LLC (asset protection) is the default. A human is the beneficial owner.
- Make the wallet company-owned. Record the wallet(s) in a Wallet Authority Schedule that binds them to the entity. The funds are the company's, not yours.
- Document the agent's authority. An Agent Mandate + Power of Attorney say what the agent may do with the wallet — so it's a delegate, not a rogue key.
- Enforce limits at runtime. A machine-readable mandate (JSON) the agent loads on every action: per-transaction and per-day caps, allowlists, prohibited categories, fail-closed if it can't verify, and a heartbeat/status check.
- Add a kill-switch + approvals. A named human (or multisig) can suspend the agent instantly; high-value or out-of-scope transfers require human approval.
- Prefer multisig custody. So no single compromised key can move everything — a hard control on top of the soft (mandate) ones.
- Keep logs. Immutable records of the agent's decisions and transactions — your audit trail.
Soft limits vs. hard limits
| Control | Type | Stops what |
|---|---|---|
| Machine-readable mandate (caps, allowlists, fail-closed) | Soft (in-agent) | Honest mistakes, drift, prompt-injection within scope |
| Multisig custody / approval gates | Hard (on-chain / human) | A compromised key or a rogue action |
| Kill-switch / revocation endpoint | Hard | Anything, immediately |
You want both. The mandate keeps a well-behaved agent inside the lines; multisig + kill-switch are what save you when something is wrong.
Where this comes from
Steps 1–7 above are exactly what an Agent Company packages: the Nevis LLC (1), a Wallet Authority Schedule (2), the Agent Mandate + Power of Attorney (3), the machine-readable mandate with fail-closed + revocation (4–5), all generated for your agent. Multisig custody, approvals, and incident response stay in your ops.
FAQ
Can an AI agent have its own wallet?
It can hold keys, but the wallet should be owned by a company — the agent operates it as a delegate.
Whose funds are in it?
The company's, if the wallet is under a company that owns the agent — which keeps it off your personal balance sheet.
How do I cap spending?
Runtime machine-readable mandate (caps/allowlists/fail-closed) + multisig + a kill-switch + approval gates.
Wrap your agent's wallet in a real entity
Form a Nevis LLC that owns your agent and its wallets, with a machine-readable mandate it obeys at runtime. Via MCP, paid in USDC. Start free.
Form your Agent Company → Run the free diagnosticKeep reading
Inside the Agent Charter The mandate, POA, and wallet schedule, in detail. AI agent compliance checklist The full list before your agent handles money. Nevis LLC for AI agents Why Nevis owns the wallet best.Informational only — not legal, tax, or security advice. Custody and key management carry real risk; review with qualified advisors. OffshoreProz is not a law firm.