What are agentic payments? They are payments initiated and completed autonomously by AI agents within the scope authorized by a user or business. In the words of Fed Governor Christopher Waller, the buyer delegates shopping and payment authority to an AI agent, which completes the transaction on their behalf. The International Monetary Fund (IMF) describes this change as a shift from “a human clicking to pay” to payments in which agents take part in decision-making.
How Are Agentic Payments Different from Ordinary Online Payments?
The difference lies in how authorization works. The IMF notes that traditional payment systems tie authorization directly to explicit instructions from a user or institution, whereas in agentic payments the agent operates under delegated authority. A single transaction may not correspond to a specific transaction instruction; instead, the user expresses goals and constraints.
To make this kind of authorization verifiable, Google’s Agent Payments Protocol (AP2) introduces the Mandate: a cryptographically signed, tamper-proof digital contract that serves as proof of the user’s instructions. Within it, the Intent Mandate sets conditions such as price limits and timing, while the Cart Mandate records the exact items and prices. Bank of England Deputy Governor Sarah Breeden has also noted that most agents are still in “recommendation mode”, with humans executing the transactions, but that requiring humans to review every action an agent takes is not realistic.
Where Are Agentic Payments Used?
The first category is machine-to-machine micropayments. Waller gives the example that, before completing a purchase, an agent may first need to pay for LLM queries, pricing data and other API calls, and such small payments are better suited to payment rails with low fixed fees. He also believes business-to-business procurement may be especially well suited to agentic commerce.
The second category is cross-border payments. The IMF notes that agentic AI can orchestrate the entire cross-border payment chain, starting from payment initiation.
The third category is consumer shopping. Card networks have launched their own solutions: Visa Intelligent Commerce uses tokenized digital credentials to confirm that an agent has been authorized by the consumer, and lets consumers set spending limits and conditions; Mastercard Agent Pay introduces Agentic Tokens and requires agents to be registered and verified.
What Role Do Stablecoins Play in Agentic Payments?
The IMF believes AI agents using stablecoin rails could become important infrastructure. Cleveland Fed President Beth Hammack has also mentioned that one of the technical advantages of stablecoins is their ability to support smart contracts and agentic workflows.
At the protocol level, AP2 supports a range of payment methods, from cards to stablecoins and real-time bank transfers, and connects to stablecoins through an x402 extension. The x402 protocol lets servers request payment directly within HTTP requests, and agents can access the resource once they have paid. However, the IMF also cautions that stablecoins have historically experienced depegging under stress, as well as reserve-related counterparty risk.
What Challenges Do Agentic Payments Face?
The first is identity and fraud prevention. The IMF notes that verifying both an agent’s identity and the intent of the user behind it becomes far more complex, and that regulatory requirements may shift from “know your customer” to “know your agent”. Waller has also said that existing anti-fraud systems are calibrated to human behavior and may not be suitable for agents.
The second is security. Both the IMF and the Financial Stability Board (FSB) list prompt injection as a risk: malicious actors may influence or even control agents by tampering with their inputs. Real-world cases of agents sending funds to the wrong place by mistake are analyzed in detail in Why Do AI Agents Accidentally Send Crypto.
The last is liability and regulation. Waller asks: if an agent buys the wrong thing, who is responsible? The IMF also notes that existing regulations struggle to distinguish between “unauthorized use” and “user negligence”, and that regulatory clarity on agent accountability and liability in payments is still lacking. Hammack, for her part, has noted that current U.S. regulations do not directly address payments made with smart contracts or agentic AI.
Conclusion
Agentic payments hand the power to initiate payments from people to AI crypto agents. The key is that the scope of authorization must be clear and verifiable, with limits and approvals that agents cannot bypass. Issues such as identity, fraud prevention and liability are still evolving. Safeheron’s MPC Wallet-as-a-Service provides businesses with MPC key management and a policy engine.
Is your business evaluating letting AI agents pay with stablecoins?
Looking for a secure way to manage digital assets?
Talk to our experts and find the right solution for your business.
Request a DemoLooking for a secure way to manage digital assets?
Talk to our experts and find the right solution for your business.