MoonPay PayBox lands in Grok, pushing agentic payments to third AI
MoonPay's PayBox, a credential vault designed to let AI agents move money on behalf of users, went live in Grok on 31 August 2026. The integration follows July launches inside Anthropic's Claude and OpenAI's ChatGPT, making Grok the third major conversational AI platform to carry the product. Taken together, the three deployments mark an inflection: PayBox is no longer a proof-of-concept but an actively expanding payments layer sitting beneath some of the world's most-used AI interfaces.
Within Grok, a user can instruct the assistant in plain language to swap stablecoins, bridge funds between blockchain networks, book a restaurant table, or purchase a flight ticket. Grok prepares the transaction; the user authorises it with a passkey; the money moves. MoonPay says no single party, including itself or the AI model, can access a user's private key independently. That claim rests on threshold cryptography and secure enclaves sourced from Sodot, a key-management firm MoonPay acquired earlier in 2026. Sodot's infrastructure is reported to secure more than $50 billion in assets across more than 10 million wallets.
Agentic payments meet traditional rails
The architecture spans both crypto and conventional payment methods. Card transactions run through Visa's agentic commerce protocol, which the company says allows PayBox to process card payments without storing or viewing raw card numbers. Supported blockchain networks include Solana, Ethereum, Base, Arbitrum, Polygon, Hyperliquid, Tempo, and Robinhood Chain. Stablecoin activity (PYUSD, USDC) is explicitly supported alongside DeFi yield strategies such as Kamino on Solana.
The underlying payment standard, x402, is positioned as an open protocol for agent-initiated transactions, meaning third-party services such as travel platforms and retailers can integrate directly rather than routing through a MoonPay marketplace. That design choice has strategic implications: it shifts PayBox from a closed fintech product towards infrastructure, analogous to how Stripe positioned itself as a rail rather than a merchant.
"AI is quickly becoming the interface through which people navigate the internet, and payments need to work wherever those conversations happen," said Ivan Soto-Wright, MoonPay's CEO and Founder.
The convergence angle: who owns the agentic transaction layer?
The race to embed payments inside AI assistants is accelerating across the fintech and blockchain sectors simultaneously, and the outcome carries significant implications for incumbent financial infrastructure. If conversational AI becomes the dominant commerce interface, the entity controlling the payment vault effectively controls the checkout layer for a material share of global retail and financial transactions. That is a position incumbent card networks, neobanks, and payments processors have historically competed fiercely to occupy.
PayBox's multi-platform strategy is a deliberate hedge. By launching across competing AI ecosystems rather than exclusively with one, MoonPay is attempting to become the cross-platform payment abstraction layer, regardless of which assistant wins user share. The self-custodial design also sidesteps the regulatory complexity of holding user funds directly, keeping PayBox in an infrastructure role while leaving custodial risk with the user.
For capital allocators, the broader question is whether agentic payment infrastructure will consolidate around a small number of open-standard layers like x402, or fragment into proprietary integrations negotiated individually between AI labs and payment networks. The Visa partnership suggests incumbents are not ceding the rail; they are embedding in it. Meanwhile, the acquisition of Sodot points to MoonPay's bet that key-management infrastructure, not the AI model itself, is the defensible moat in this stack. Cross-sector investors watching both the AI platform race and the digital-asset custody market should treat these two convergence vectors as increasingly inseparable.