As artificial intelligence continues its rapid evolution, 2026 is shaping up to be a pivotal year for agentic payments: payment transactions initiated, managed, and executed by adaptive AI systems that can act autonomously, with delegated authority to act on behalf of a user. While automation in payments has existed for a long time (think “autopay” for recurring payments), agentic AI takes automation to another level by making decisions and taking actions to achieve goals defined by the user. This has the potential to streamline and optimize a number of different activities, like shopping, financial planning, and investing. Agentic AI may even create entirely new ways of executing payments, both in fiat currency and digital assets. The business use cases are already evident and ever increasing, and with them, a growing wave of commercial interest in optimizing every conceivable area of financial intermediation. But agentic payments also raise questions (about regulatory compliance, consumer protection, security risks, and, generally, who pays when the agents fail) to which businesses and regulators may need to respond as the technology matures and becomes more widely available.
Even before legal and regulatory questions are resolved, there are practical hurdles to overcome for agentic payments to be adopted at scale. A key hurdle is whether one dominant protocol (or multiple interoperable protocols) will emerge across the industry to permit the interoperability of autonomous AI agents, such as Anthropic’s Model Context Protocol (MCP) or Google’s Agent2Agent Protocol (A2A). A shared protocol is also needed to enable AI agents to make, process, and receive payments and to solve related issues that the current infrastructure for digital payments is not necessarily well suited to address, including how to verify that a user gave the AI agent authority to make a particular payment. Several open payment protocols designed, in part, to address these issues have been announced recently, including the Agent Payments Protocol (AP2) led by Google, the x402 extension on A2A developed by Coinbase (creating a crypto-native agentic payments model), and the Machine Payments Protocol (MPP) co-authored by Stripe and Tempo. Established payment networks (such as Mastercard and Visa) have also announced their intent to adopt agentic payments while leveraging their existing authentication and security standards.
At the same time, at the intersection of crypto and AI, stablecoins and other digital assets promise to play a significant role in the rollout of agentic payments, as illustrated by some of the payment protocols referenced above. Unlike traditional payment rails (e.g., ACH transfers and credit cards), crypto-based payments are initiated by the sender (i.e., a “push system”), which intrinsically empowers the sender to leverage additional crypto-specific advantages, such as frictionless cross-border settlement, 24/7 availability, ease of microtransactions, and relative ease of foreign currency exchange. Depending on the nature of the protocol, certain safeguards can also be layered into crypto transactions, allowing developers to set spending limits and otherwise throttle the velocity of transactions directly in code.
These developments are occurring against an architectural backdrop of federal and state laws and regulations that do not provide clear answers to questions arising in the context of agentic payments. For example, who should be responsible if an incorrect or fraudulent payment is made through an agentic AI system: the user, the developer of the AI system, the entity making the AI system available to the user (e.g., the bank or the payment processor, which may not necessarily be the developer), or the merchant? Below, we outline three key legal and regulatory challenges that companies should keep in mind if they intend to offer or incorporate agentic payments into their service offerings.
Agentic payments are likely to be governed by a layered legal and regulatory framework that combines AI-specific and financial services laws and regulations, in addition to general-purpose laws. State AI-specific laws continue to emerge, and the broader AI regulatory environment in the United States is still being defined. Building on a December 2025 executive order, the Trump administration recently released the National Policy Framework for Artificial Intelligence Legislative Recommendations, outlining recommendations for Congress to establish a federal AI policy framework and prevent a fragmented patchwork of state regulations. While existing state laws and the main discussion points at the federal level are not specifically oriented toward agentic AI or the use of agentic AI systems for payment purposes, companies operating in this space should consider monitoring the evolving legislative landscape and planning for a period of legal uncertainty amid ongoing state-level activity and potential federal challenges to certain laws. Questions also remain as to how courts will apply general-purpose laws (such as those governing product liability, consumer protection, and agency relationships) to AI. This evolving legal and regulatory environment may affect how and when agentic payment services are released to the broader public.
The existing U.S. money transmission legal and regulatory regime was built around human intermediaries holding or moving funds on behalf of others through a system of actual or constructive control over third-party funds. In the United States, money transmission is regulated primarily at the state level, requiring licensure in every state where a company operates, plus federal registration as a money services business (MSB) with the Financial Crimes Enforcement Network (FinCEN). The trigger for licensure turns on whether an entity is “receiving” money for transmission or exercising meaningful control over the movement of funds on behalf of another party. Applied rigorously, this standard could implicate a number of agentic platform operators if their systems are designed to maintain such control, even where the design is intended to prevent fraud, risk of loss, and other potentially harmful outcomes. The key fault line is not whether a system is technically “autonomous” but whether the platform underlying that system has retained meaningful authority over payment flows.
On one end of the spectrum, an agentic tool that merely executes pre-authorized, user-directed transactions with no platform-level ability to redirect, withhold, delay, or otherwise manage funds is more likely to be deemed a pure technology conduit, analogous to the software tools long recognized as outside the definition of money transmission. The protocols described above attempt to enable this type of tool by providing a method of establishing clear, demonstrable consent (and, therefore, authorization to the transaction) through code-based signatures. At the other end, a platform that holds or co-holds wallet keys; maintains an override capability over disbursements; can pause or redirect transactions in response to its own risk logic; or pools user funds at any point in the flow may be subject to heightened regulatory scrutiny (including licensing requirements), regardless of whether the front-end product is marketed as an AI assistant, a shopping agent, or a financial management tool.
The need to obtain money transmission licenses and comply with applicable regulation may add significant complexity to the deployment of agentic payment systems, and it is currently unclear how these issues will be viewed and resolved by federal and state authorities.
The third legal challenge (and perhaps the most immediate one) comes from the implementation of the Electronic Fund Transfer Act (EFTA) and related Regulation E in the context of agentic AI. Regulation E’s consumer protections hinge on the foundational distinction between whether a transfer was “authorized” or “unauthorized.” An unauthorized electronic fund transfer (EFT) is defined as one “initiated by a person other than the consumer without actual authority to initiate such transfer,” which itself often turns on establishing “demonstrable consent.” It is currently unresolved whether a consumer granting an AI agent access to their bank account or payment credentials satisfies Regulation E’s authorization requirements. Moreover, while Regulation E states that consumers can authorize a payment using a “card, code, or other means,” it leaves open what happens if an AI agent violates the consumer’s instructions. Regulation E is also generally understood to apply to EFTs involving consumer deposit accounts at financial institutions, and its protections have not been extended to crypto-native payment rails. This may create an additional layer of uncertainty since hybrid models that support both fiat and digital assets are likely to be common in agentic payment systems, including by enabling stablecoins to be minted through transfers from a consumer’s depository account.
The lack of a clear framework for handling disputes in agentic commerce and payments means that consumers could be left without standard dispute rights when contesting agentic purchases, regardless of where the error occurred. Again, emerging agentic payment protocols attempt to resolve this problem. AP2, for example, uses cryptographically signed “mandates” that capture the user’s upfront scoped instructions and final approval of a specific purchase to establish and create an audit trail of user intent. However, this type of code-based trail of authorization is still untested in courts and before regulators.