Coinbase Agentic Wallets let AI agents hold and spend onchain
Coinbase is rolling out Agentic Wallets, a wallet infrastructure built for autonomous AI agents to hold funds and transact directly on public blockchains, as reported by The Block. The launch frames onchain accounts as capabilities that can be provisioned to software agents rather than just to human end users.
The initiative relies on the x402 protocol so agents can initiate and complete onchain transactions natively, as reported by CryptoBriefing. In practical terms, this pairs an agent’s decisioning logic with an execution rail that supports machine-to-machine payments and other programmatic financial tasks.
Agentic Wallets are presented alongside controls that limit what an agent can do, including per-session caps, transaction-size caps, and customizable restrictions, according to PYMNTS. The design goal is to equip an agent with spending power quickly while constraining scope and requiring explicit approvals for sensitive actions.
Why Payments MCP gating and x402 protocol matter for safety
Model Context Protocol (MCP) sits between an AI model’s reasoning and a wallet’s signing authority, so an agent can propose payments or check balances but cannot unilaterally move funds; that gatekeeping function is central to safety, said Sean Ren, co-founder at Sahara AI. In this setup, users retain the final authorization step while the model’s suggested actions flow through a controlled interface.
Even with that separation, risk management remains essential. Practical risks include prompt injection that alters instructions, exposure to malicious or fragile smart contracts, man-in-the-middle threats in tool chains, and poor handling of slippage or liquidity when swapping assets.
Editorially, the core trade-off is adding productive autonomy without removing human accountability. “Giving AI agents wallet access means introducing a layer of trust into what is supposed to be trustless,” said Aaron Ratcliff, Head of Intelligence at Merkle Science.
On the trust and reputation layer, enterprises are already exploring ways to make agent identities verifiable and legible to humans, according to Forbes. That work complements gating by attaching recognizable identifiers and attestations to the agent that is proposing or executing financial actions.
Immediate impact for users, devs, and enterprises right now
For individual users and developers, the near-term pattern is to start with constrained scopes: set spend and session limits, allowlist specific contracts and tokens, route sensitive actions for human approval, enable detailed logging, and verify behavior on testnets before moving to mainnets. This is a governance problem as much as it is a technical one, and the controls noted above are designed to be adjusted over time as confidence grows.
In terms of what agents can do today, basic DeFi operations such as sending, swapping, and lending are a reasonable first wave while more complex portfolio workflows develop, according to Brian Huang, CEO of Glider. Early autonomy tends to work best where data is structured, counterparties are known, and failure modes can be capped with small, pre-approved spend limits.
For enterprises, x402-backed automation points toward recurring and machine-to-machine payments, inventory or subscription reconciliations, and operational disbursements with audit trails. The combination of MCP gating, scoped wallets, and verifiable activity logs can make onchain tasks fit existing control frameworks rather than bypass them.
At the time of this writing, Coinbase Global, Inc. (COIN) was at $165.86 (+0.45%) in after-hours trading, with a 52-week range of $142.58 to $444.65 and a beta of 3.70, based on data from Nasdaq. Those figures provide market context for the rollout but do not bear directly on the technical or governance properties of Agentic Wallets.
Security model, risks, and Coinbase safeguards to know
The security model depends on two layers: proposal and permission. Payments MCP provides the proposal layer where an agent can request actions within predefined scopes; Agentic Wallets enforce permission with spend caps, per-session limits, allowlists, and optional human-in-the-loop approvals. In combination, these controls reduce blast radius if an agent is misled or interacts with an unsafe contract.
Identity and compliance are converging on cryptographic credentials that prove who or what initiated a transaction. Research communities are evaluating decentralized identifiers and verifiable credentials to bind actions to accountable entities without exposing unnecessary data, according to arXiv. In practice, this supports auditability, reduces impersonation risk, and can align agent activity with enterprise policy.
It is prudent to assume that autonomy will surface edge cases. Teams should validate data sources, cap slippage for swaps, and maintain real-time monitoring for anomalous patterns, recognizing that controls mitigate but do not eliminate risk. The safeguards described here are designed to be configurable, so organizations can calibrate them to their risk appetite and regional obligations.
| Disclaimer: This website provides information only and is not financial advice. Cryptocurrency investments are risky. We do not guarantee accuracy and are not liable for losses. Conduct your own research before investing. |
