Who Controls an Agent Wallet's Keys?

An agent needs to pay $1 for data. A wallet can provide an address and a signature, but the questions that shape risk are: who can approve a transfer, and who can change that rule?

Scope: This article summarizes technical and product arrangements. It does not assess a provider’s security, and no wallet integration or funds test was performed.

flowchart LR A[User or company sets rules] --> B[Agent requests signature] B --> C[Wallet or signing system checks policy] C --> D[Approve and sign, or reject]

Four terms describe different things

Term	What it primarily describes	What to ask next
Custody	A provider has an agreed role in controlling assets or managing keys	What can the user and provider each do? What happens if the service stops or freezes access?
Non-custody	Usually emphasizes some degree of user control	Which key and which permissions? Can the provider transfer assets alone?
MPC	A cryptographic approach in which multiple parties participate in key generation or signing	Who holds shares? What are the signing threshold, recovery and policy settings?
Smart account	An account whose behavior is controlled by code and configuration	Who can upgrade rules? Which actions are restricted? What are the contract and network risks?

These categories are not mutually exclusive: one product can combine MPC and a smart account. MPC does not automatically mean the user has ultimate control. The threshold, share holders and recovery rules matter. Ethereum’s account abstraction overview describes programmable account security; implementations differ.

“The agent cannot see the private key” is not enough

An agent may still hold an API credential that can submit signing requests. If the service allows arbitrary recipients without amount limits, a leaked credential can still lead to loss. A signing system may reject requests outside its policies, but only a review of the policies and covered actions reveals what they protect.

Check separately where keys or shares live, who can request a signature, who can change policies, which actions bypass them, and how to recover and revoke access. For smart accounts, also examine upgrade permissions and module risks.

Questions for a provider

Can the user, agent, platform or another participant transfer funds alone?
What happens to signed or submitted transactions after access is revoked?
How are funds recovered if a provider becomes unavailable or a device is lost?
Do the same policies apply across networks, assets and transaction types?
Can the user see and receive notice of policy changes?

Continue with agent payment security and agent wallet basics.