Only $1.1 million in transaction volume has passed through Coinbase’s x402 payment protocol in the past 30 days. That number tells a lot about where AI-driven crypto payments actually stand right now.
A New Layer For Crypto Transactions
Base, the Ethereum layer-2 network operated by Coinbase, has rolled out a tool that lets AI agents carry out blockchain operations directly from a chat interface.
Called Base MCP, the system works with AI models including Anthropic’s Claude and OpenAI’s ChatGPT, allowing users to transfer funds, swap tokens, check balances, and pull up transaction history without leaving the conversation window.
The tool also connects to a range of crypto apps, including Morpho, Moonwell, Uniswap, Aerodrome, Avantis, Bankr, and Virtuals.
Users interact with these platforms through the AI agent, which proposes actions that must then be approved by the user through a separate wallet window. The agent has no access to private keys.
Confirmation Required Before Funds Move
Lincoln Murr, Coinbase’s head of AI Product, described Base MCP as a wrapper on top of existing APIs.
He told Fortune that unlike standalone agentic wallets confined to a terminal, the Base account syncs across both in-agent and in-app activity, carrying a user’s trade history and portfolio wherever they go.
Base said every proposed transaction goes through the same review process users see with standard Base account requests. Asset changes are simulated before the user confirms anything. No funds move without a deliberate approval step.
The launch is also expected to drive more activity through x402, the agentic payment standard Coinbase introduced in May 2025.
The protocol is designed to let AI agents make small crypto payments as part of a broader micro-transaction economy.
Data shows that economy is still getting started — x402 has processed just $1.1 million in volume over the last 30 days, according to x402scan.
Security Researchers Flag Risks
Not everyone is convinced the infrastructure is ready for wider use. A research paper from Google and several universities concluded that AI agents should be treated as untrusted components within any system.
The research warns that bad actors can manipulate agents by embedding hidden instructions inside data the agent processes.
Those concerns are not purely theoretical. The developer platform Socket recently uncovered malware targeting crypto developers that worked by injecting concealed instructions into AI coding tools to redirect their behavior.
Reports indicate that Base MCP supporters argue the confirmation step addresses the core risk, but researchers say the problem runs deeper than any single safeguard can fix.
Featured image from Coinbase, chart from TradingView
