MoonPay announced the launch of a non-custodial payment vault PayBox, allowing AI assistants (such as OpenAI's ChatGPT and Anthropic's Claude) to prepare and execute cryptocurrency transactions and online purchases on behalf of users. After users connect PayBox to a supported AI platform, they can describe their needs in natural language, and the AI assistant can search options, prepare transactions, and complete payments within the permissions set by the user.
PayBox is not limited to crypto-native transactions. It supports fiat purchases of digital assets, token swaps, cross-chain transfers, depositing funds into DeFi protocols, and more; it can also book flights, restaurant seats, and purchase goods from online retailers depending on integrations. In essence, the AI assistant becomes a payment interface, allowing users to complete multiple financial and commercial steps through a single conversation.
Addressing the most critical trust issue in AI agent payments, PayBox designs two authorization modes: in "Always Ask" mode, every transaction requires user authorization via passkey, and the authorization cannot be reused; "Autonomous" mode allows the AI to operate automatically within user-set spending limits and rules, without requiring confirmation for each transaction. If users want to modify the authorization mode or transaction rules, they must provide passkey verification again.
On the technical foundation, PayBox's wallet transactions use multi-party computation (MPC) technology, which shards private keys across multiple secure environments, so neither MoonPay nor the connected AI system can independently reconstruct the full private key and authorize transfers. Card payments use Visa's Agentic Commerce technology, enabling AI to complete approved transactions without touching card numbers.
PayBox supports Solana and multiple EVM-compatible networks, including Ethereum, Base, Arbitrum, Polygon, Hyperliquid, Tempo, and Robinhood Chain. Additionally, MoonPay integrates the x402 payment standard, which is specifically designed for servers to receive payments initiated by AI agents.
MoonPay is not the only company betting on this direction. Coinbase is also advancing AI agent interaction with crypto wallets, launching the Coinbase for Agents program and developing the AgentKit tool, allowing developers to give AI agents access to crypto wallets, execute transactions, and make payments; its x402 technology also aims to streamline payment flows between AI agents and online services.
Robinhood is also opening up parts of its financial infrastructure to AI agents, allowing them to trade and purchase financial products under security controls and limits. This means that in the future, users may no longer need to manually open apps, search for assets, and confirm orders, but instead tell the AI the desired outcome, and it will execute the steps within set boundaries.
Traditional payment giants are also joining the fray. Visa launched Visa Intelligent Commerce, which supports AI agents in discovering products, comparing options, and completing purchases on behalf of consumers, and is developing tokenization, identity authentication, and spending control capabilities, while also exploring the combination of stablecoins with AI. Mastercard launched Agent Pay for Machines, partnering with Coinbase, OKX, Tempo, Stripe, and others, and has already demonstrated AI agent payments in collaboration with Santander Bank, showing that the technology is moving from crypto experiments to regulated financial environments.
Progress on the AI side is equally critical. OpenAI is pushing ChatGPT from information retrieval and product discovery to real transactions, through Instant Checkout and the Agentic Commerce Protocol developed with Stripe, allowing users to purchase products directly within ChatGPT. Initial partners include Etsy merchants, with plans to expand to more Shopify merchants such as Glossier, SKIMS, Spanx, Vuori, and others.
Stablecoins are expected to become an important settlement rail for AI agent payments. Compared to traditional payment systems that rely on bank business hours, geographic boundaries, and multiple intermediaries, blockchain assets can move across networks 24/7, making them particularly suitable for small, high-frequency, automated, or cross-border agent-to-agent payments. Visa and Mastercard are both exploring hybrid models that use both traditional payment networks and stablecoins, suggesting that the future may not be crypto replacing bank cards, but rather cards, stablecoins, and blockchain networks together forming different payment rails for AI-driven commerce.
A deeper change is that the interface through which people interact with money is changing. When users can directly ask an AI agent to "book a flight within a budget" or "buy a certain product," and the AI handles searching, decision-making, and payment, the presence of wallets or payment apps themselves will diminish, and competition will shift to the infrastructure layer where AI agents securely access, move, and settle funds. This is the opportunity for crypto and stablecoins: they are not just another payment method, but programmable assets that can flow across blockchain networks 24/7.
The winner of this race may not be the best crypto wallet or payment app, but the infrastructure provider that enables AI agents to securely and autonomously access, transfer, and settle funds.
