AI agent wallets gain traction as users split on trust

MetaMask and Coinbase now ship AI agent wallets, while readers remain split on trusting software with funds.

MetaMask and Coinbase have moved AI-agent wallet control from theory to shipping products, and a Cryptopolitan poll shows the audience is still divided on whether that is acceptable. The debate is no longer about whether the tooling exists; it is about who should trust it, under what rules, and with how much money.

What changed

On June 8, Consensys-owned MetaMask opened early access to Agent Wallet, a self-custodial wallet designed for AI agents to execute swaps, perpetual trades, prediction-market bets, and liquidity moves across nine EVM chains and Hyperliquid. Users keep the keys and can set spend caps, allowlists, and 2FA requirements for actions outside preset rules.

MetaMask says every transaction will be threat-scanned, and safe transactions are covered up to $10,000. The company is not alone: Coinbase launched Agentic Wallets in February, calling it wallet infrastructure for non-human operators, with keys isolated inside hardware-backed Trusted Execution Environments.

• Coinbase says x402 has crossed 480,000 active agents.
• The protocol has processed more than 165 million transactions.
• Total cumulative volume has topped $50 million.
• Weekly transactions rose from about 548,000 to 2 million in three months.
• MoonPay has integrated Ledger wallets for AI transactions.
• The x402 Foundation counts Google, Visa, AWS, Circle, Anthropic, Stripe, Cloudflare, and the Solana Foundation among its backers.

Other players are moving too. MoonPay integrated Ledger wallets for AI transactions and backed the Open Wallet Standard alongside PayPal and the Ethereum Foundation. Cobo has shipped an Agentic Wallet using MPC security, while the x402 Foundation now sits under the Linux Foundation umbrella.

Why it matters

The practical shift is simple: AI agents can now watch markets nonstop, fire trades when conditions match, route orders to cheaper venues, rebalance positions, and claim rewards without waiting for a human to log in. That matters in crypto because the market never closes, and human operators cannot keep pace with 24/7 execution.

The risk is just as clear. If an agent hallucinates, gets hacked, or misfires on a trade, the user still bears the loss. That is why the trust model is now the real product question, not the wallet itself.

Cryptopolitan’s poll reflects that tension. The leading answer was “Nope” at 30.22%, but nearly 70% of respondents said yes in some form if the brand is trusted or guardrails are in place. The market is not rejecting agent wallets outright; it is asking who gets to hold the signing flow.

That makes agent wallets a test case for the wider “agentic economy” that Coinbase CEO Brian Armstrong says is already forming on Base. If machines are going to pay machines, the rails, insurance, and controls need to be settled before adoption scales further.

The real story is not whether AI wallets exist anymore. It is whether users, exchanges, and infrastructure providers can agree on a trust model before autonomous spending becomes normal.