MoonPay launched MoonPay Agents as a non-custodial software layer that lets AI systems create and control wallets, fund them through fiat rails, and execute on-chain transactions with pre-built “financial skills.” The key shift is that agentic AI moves from analytics to direct economic action, turning model outputs into settled transactions on public blockchains.
The release is positioned to scale from thousands to potentially millions of autonomous agents while keeping private keys with the user and adding programmable guardrails. For developers, compliance teams, and institutional users, the operational takeaway is that automated decisioning is now being productized into executable cash flow with policy controls baked into the workflow.
Your AI agents need crypto. Now they have it.
Introducing MoonPay Agents, the non-custodial infrastructure for autonomous transactions.
➡️ Onramp
🔀 Swap across chains
🔁 Trade onchain
⬅️ OfframpMulti-chain. Native x402 support. One CLI prompt.
🤖 https://t.co/7Shff929Og pic.twitter.com/zlTYlns1tW
— MoonPay 🟣 (@moonpay) February 24, 2026
How the product is built and how it plugs in
MoonPay Agents is delivered through a developer-facing CLI and API that exposes wallet lifecycle operations such as creation, funding, and programmatic transaction execution. This design makes agents easy to embed into scripts, schedulers, and machine-to-machine pipelines, which materially lowers the friction to automate on-chain activity at scale.
The platform integrates MoonPay’s existing fiat and stablecoin rails, including virtual accounts in U.S., EU, and GBP currencies, recurring buys, and real-time swaps. MoonPay describes the stack as non-custodial by default: the user retains private keys while granting an AI agent scoped permissions that are constrained by explicit limits.
MoonPay also positioned the product as an extension of infrastructure already used by nearly 500 enterprise customers and a user base described as more than 30 million. That distribution footprint matters because it creates a credible path for agent flows to become high-frequency and widely distributed rather than a niche developer experiment.
Guardrails, KYC, and why those controls exist
MoonPay’s control model is built around two pillars: a one-time KYC process for the human user and programmable transaction policy for the agent. The intent is to establish a regulatory envelope for the user, then restrict what an agent can do inside that envelope through enforceable permissioning.
The controls described include spending caps, transaction-type restrictions, address whitelisting, and multi-factor approval for higher-risk actions, plus the ability to constrain agents to specific activities like swaps or recurring stablecoin payments rather than arbitrary transfers. These constraints are meant to reduce the blast radius of mistakes or compromised agent behavior in an environment where transactions are irreversible and liability for autonomous agents remains unclear.
MoonPay’s own documentation highlights protocol and implementation risks, including bugs or vulnerabilities that could impair functionality or lead to asset loss, and frames KYC-plus-permissions as the primary compliance lever. In governance terms, the product is trying to convert “AI autonomy” into “bounded autonomy” that can be audited and defended.
What changes if agent adoption scales
Operational features cited include automated token swapping, portfolio rebalancing, recurring stablecoin payments, and algorithmic arbitrage execution, which collectively shorten the loop from signal to settlement. If agent usage expands meaningfully, on-chain activity could shift from human-initiated transactions to persistent programmatic flows with different timing, volume, and behavioral patterns.
At scale, that implies higher transaction entropy, more unique daily actors, and altered gas distribution, which can change network load profiles and execution conditions for everyone else. It also increases the likelihood of regulatory attention centered on liability, KYC consistency, and whether programmable limits are genuinely enforced in production.
For product teams and compliance officers, the near-term workstream is measurable and execution-heavy: define and monitor agent transaction profiles, model gas and block-density impact, and convert permission schemas into auditable policy that can withstand review. The central challenge is ensuring that “non-custodial” does not become “uncontrolled” once autonomous flows reach operational scale.
