MoonPay Launches PayBox to Embed a Secure Crypto Wallet Inside Claude and ChatGPT, Enabling AI-Driven Payments
Table of Contents
You might want to know
Can AI chatbots like Claude and ChatGPT not only plan purchases but actually complete payments securely on your behalf?
How does PayBox protect users’ private keys and card details so neither MoonPay nor the AI agent can misuse funds?
Main Topic
MoonPay has introduced PayBox, a payment vault that integrates directly into AI chatbots such as Claude and ChatGPT, enabling AI agents—autonomous software designed to perform tasks for users—to execute financial transactions. The PayBox vault functions as a secure container that can hold both cryptocurrency wallets and traditional payment cards. When a user asks the chatbot to make a purchase, the AI prepares the transaction and the user confirms it using a passkey stored on their device (for example, a biometric or device-based approval like Face ID). Once approved, the funds are transferred to complete the transaction.
Before PayBox, chatbots could assist with research, price comparisons, and shopping recommendations, but they required users to leave the conversation, open a browser, and manually complete payments. PayBox removes that friction by allowing the agent to carry out the transaction flow within the chat environment while retaining user consent and control over approvals.
At launch, PayBox supports integrations that let users make restaurant reservations, book flights, and shop at major online retailers through agent-assisted flows. On the crypto side, the product also allows users to swap tokens, bridge assets across blockchains, and route funds into decentralized finance (DeFi) protocols such as Aave. The system launched with support for Solana plus seven Ethereum-compatible chains: Ethereum, Hyperliquid, Tempo, Base, Robinhood Chain, Arbitrum, and Polygon. PayBox is also compatible with x402, an open payment protocol initiated by Coinbase and later donated to the Linux Foundation, which is now supported by companies including Visa, Mastercard, Stripe, Shopify, AWS, and Anthropic. x402 enables AI agents to pay services built to accept agent-initiated transactions.
The technical foundation for PayBox comes from Sodot, an Israeli key management startup that MoonPay acquired in April for roughly $100 million in an all-stock deal. Sodot’s technologies already protect more than $50 billion in assets and secure over 10 million wallets for clients such as eToro and BitGo. MoonPay’s development toward PayBox followed several prior moves during the year: in March, it released an open wallet standard for AI agents backed by organizations including PayPal, the Ethereum Foundation, and the Solana Foundation; in May, it launched an in-chat app that let users buy crypto inside ChatGPT (though that earlier version still redirected users to a checkout page); and in June it shipped a desktop app linking Claude and Codex directly to crypto wallets.
Security is central to PayBox’s design. Wallet private keys are protected using multi-party computation (MPC), a cryptographic technique that splits a private key into multiple fragments held by different parties so that no single entity can reconstruct or use the full key alone. Those key fragments are stored inside hardware-isolated trusted execution environments (TEEs), which are hardware-backed enclaves that protect sensitive operations even from the host server. According to MoonPay, wallet keys are split via MPC across hardware-isolated enclaves, so no device, session, or party—not MoonPay, not the AI agent—ever holds a complete key or can sign alone.
For card payments, PayBox uses Visa’s agentic commerce protocol, allowing the AI to process payments without exposing the user’s full card number to the agent. That design reduces the risk of card credential leakage and reuse, as the AI never sees raw card details during the transaction flow.
MoonPay frames PayBox as a resolution to a longstanding tradeoff between convenience and control. Many existing solutions require users to grant custody of funds to a single platform, hosted wallet, or agent to enable autonomous transactions—introducing a concentration of trust and risk. PayBox is intended to provide convenience while maintaining control by ensuring that no single party, including MoonPay and the AI itself, can unilaterally access or move a user’s funds or credentials.
Key Insights Table
| Aspect | Description |
|---|---|
| Product | PayBox: a payment vault embedded in chatbots that stores crypto wallets and payment cards. |
| Primary Function | Enables AI agents to initiate and complete payments with user approval via passkeys. |
| Security Mechanisms | MPC-split keys stored in TEEs; no single party can sign transactions alone. |
| Blockchain Support | Solana and seven Ethereum-compatible chains, including Ethereum, Arbitrum, and Polygon. |
| Card Payments | Routes through Visa’s agentic commerce protocol so agents never see full card numbers. |
| Underlying Tech | Built on Sodot’s key management technology, acquired by MoonPay earlier in the year. |
| Use Cases at Launch | Restaurant bookings, flight reservations, retail shopping, token swaps, bridges, and DeFi access. |
Afterwards...
As AI agents proliferate, the ability for them to hold and move money securely will become a foundational capability for many consumer and enterprise services. Continued exploration is needed in areas such as cryptographic key management (including advances in MPC and TEE robustness), agent-authentication standards for cross-service payments, and interoperable open protocols like x402 that reduce friction for agent-initiated commerce.
Further research into adversarial threats against TEEs and MPC implementations, improved transparency and auditability of agent actions, and standards for user consent and recoverability will help balance innovation with safety. Equally important are policy and regulatory frameworks that clarify responsibilities and liabilities when autonomous agents transact on users’ behalf.
In short, PayBox represents a step toward conversational money—where payments become part of an ongoing dialogue with AI—while emphasizing cryptographic techniques and hardware protections to preserve user control. The next phase of development should focus on strengthening technical safeguards, interoperability, and governance to make agent-assisted payments broadly trustworthy and practical.