MetaMask has launched an early access version of Agent Wallet, a new non-custodial product that allows AI agents to execute crypto transactions under user-defined controls across Ethereum-compatible networks and Hyperliquid.
Summary
- MetaMask has launched Agent Wallet, allowing AI agents to execute trades, swaps, and other on-chain transactions under user-defined controls.
- Every transaction is simulated and screened by Blockaid, with risky activity requiring additional user approval before execution.
According to a press release shared with crypto.news, the wallet is designed for autonomous agents that can carry out tasks such as token swaps, perpetual futures trading, liquidity provisioning, and prediction market activity without requiring manual input for every step.
The launch places MetaMask among a growing list of crypto companies building tools for AI-powered finance. Over recent months, firms across the sector have introduced products that let AI systems interact with wallets, trading platforms, and payment networks while keeping final authority with human users.
Joe Lubin, founder of Consensys, said crypto infrastructure is well suited for machine-driven transactions because autonomous software can coordinate and verify activity through blockchain networks.
“Agents will manage real capital and make real financial decisions, and the infrastructure underneath has to be worthy of that. MetaMask Agent Wallet is the first agent wallet built with comprehensive full stack security for that world: one where agents act with autonomy, security is mandatory, and the person behind the agent stays in control.” – Joe Lubin, Founder and CEO of Consensys and Co-Founder of Ethereum.
Security controls remain central to rollout
Built around preset permissions, Agent Wallet requires users to establish spending limits, approved transaction lists, and other operating rules before an AI agent can access funds, according to Consensys.
Every transaction must first pass through a simulation process that checks the expected outcome before execution. Consensys said the wallet also integrates security monitoring from Blockaid, which scans transactions for potential scams and suspicious activity.
Where a transaction falls outside a user’s predefined rules or is considered risky, Blockaid can trigger a two-factor authentication request through email or push notification before execution proceeds.
The announcement further added that MetaMask’s Transaction Protection program will cover eligible transactions deemed safe by the platform for up to $10,000, subject to applicable terms and conditions.
AI wallet race gains momentum
Support for Agent Wallet extends across several AI development environments, including OpenClaw, OpenAI Codex, Claude Code, Cursor, and Nous Research Hermes Agent. Consensys said the product is framework agnostic and can work with different agent architectures.
For now, access is limited to a small group of users through a command-line interface as part of an Early Access Program. A wider rollout is expected later this summer, according to the company.
Elsewhere in the industry, crypto firms have increasingly connected wallets and payment systems to AI agents. Gemini has introduced tools that let users connect AI trading bots to exchange accounts, while card issuers and wallet providers have explored dedicated financial accounts for autonomous software.
A similar approach emerged in May when Base introduced Base MCP, a system that connects AI agents with Base Accounts. According to Base, the tool lets users conduct transfers, swaps, portfolio tracking, and other onchain actions through chat interfaces while requiring explicit user approval before any transaction is signed.
Base said its MCP system supports ChatGPT, Claude, Codex, and Cursor, and integrates with decentralized finance applications including Uniswap, Morpho, Moonwell, Aerodrome, and Avantis. The company maintained that private keys remain inaccessible to AI agents, with transaction approval staying under user control.
Security concerns have continued to accompany the rise of agent-based crypto products. In a recent report, researchers from Google, Meta, Gray Swan AI, EmbraceTheRed, and several universities argued that AI agents should be treated as untrusted components and separated from sensitive systems and instructions.
