Product Launch2 min read

AWS Lets AI Agents Spend Money on Your Behalf

June 5, 2026Synthesized from 1 source: AWS

Amazon launched a preview service that lets AI agents make real purchases autonomously, with built-in spending limits and guardrails, signaling that software working on your behalf will soon be able to pay for things without asking you first.

Amazon launched a preview of a payments service inside its AWS cloud platform that lets AI agents make purchases without waiting for a human to approve each one. The service, called Amazon Bedrock AgentCore Payments, was built with Coinbase and Stripe. It is currently available in preview across four regions: the US East Coast, US West Coast, Frankfurt, and Sydney.

The problem it solves is real and already affecting teams that use AI agents for research or operations. When an agent running a task hits a resource that costs money, it currently stops or falls back to older, free data. A financial analyst's agent that needs live pricing data from a paid feed, or a procurement agent that needs to query a supplier's paid portal, gets stuck at that wall. This service lets the agent pay and keep going.

The payments are not open-ended. Developers set a spending cap and an expiry time before the agent starts working. If the agent tries to spend beyond the cap, the transaction is rejected automatically. That rejection happens at the infrastructure level, not inside the AI itself, which matters: it means the limit cannot be talked around, no matter what instructions someone feeds the agent.

Card numbers and banking details never touch the agent at all. The end user funds a separate digital wallet through Coinbase or Stripe's own portals, then explicitly grants the agent permission to spend from it. The agent only sees a scoped permission to draw from that wallet, not the underlying payment details. This keeps the compliance burden with the wallet providers, not with whoever built the agent.

Every transaction is logged automatically. Finance, security, and compliance teams can see what the agent paid for, how much, and when, without relying on the agent to report its own activity. That audit trail matters as soon as regulators, auditors, or internal controls ask what a system spent and why.

AWS is not alone here. Google has its own agent payment protocol, and Stripe has been rebuilding its infrastructure specifically for machine-to-machine commerce. The convergence is fast: the x402 payment protocol that this service runs on processed over 169 million payments in roughly its first year, across more than 590,000 buyers and 100,000 sellers.

For operators in industries like insurance, logistics, travel, or financial services, the near-term implication is concrete. Agents that currently handle research, procurement queries, or data gathering are about to gain the ability to pay for premium sources and services on their own. That is useful, but it also means spending policies need to exist before those agents are deployed, not after. The audit logs this service produces automatically are exactly what finance and compliance teams will eventually ask for.

The service is in preview and the APIs may still change before full release. Companies already on AWS can access it without new infrastructure. Those not on AWS should note that Google and others are building competing versions, so this capability is arriving across platforms, not just through Amazon.

Stay informed

Get AI intelligence like this delivered to your inbox.


You May Also Find Valuable