Fed’s Waller Says AI Shopping Changes What Banks Need to Authenticate

Federal Reserve Governor Christopher Waller highlighted Tuesday that AI-driven commerce introduces fresh challenges for banks and payment systems in terms of authentication. In his Sibos speech, Waller emphasized the need to prove an AI system's authority to execute payments on behalf of consumers or businesses, rather than just assisting them. This shift in authentication requirements could prove to be a critical infrastructure hurdle for agentic commerce.

Visa, Mastercard, and PayPal have already begun developing technologies enabling software agents to participate in shopping and payment processes. However, the banks assessing these transactions may require more than just a valid payment credential. They may also need proof that the consumer or company has explicitly authorized that specific agent to act within predefined limits.

This situation could give rise to a novel payment object: a machine-readable record of delegated authority, akin to a digital power of attorney. This record could specify the authorized agent, the scope of transactions, the payment method, and the authority's validity period.

Several industry entities are already progressing towards this direction. For instance, Visa's Intelligent Commerce technology can link payment credentials to a specific agent and enforce controls ensuring purchases align with the user's instructions. Mastercard's Agent Pay system includes registered agents and "verifiable intent," intended to validate user consent before an agent acts.

Waller did not propose a uniform standard but identified three primary challenges: authentication, liability, and fraud. Existing fraud systems are tailored around human behavior and may need to be adapted for software agents. Additionally, Waller pointed to the necessity of technical standards that could capture the intended transaction and how agents carry out those instructions.

The stakes may be particularly high in business-to-business (B2B) payments. B2B transactions are often governed by rules such as approved suppliers and budget limits, making them more suitable for agents. These agents could potentially negotiate terms and select payment strategies to optimize working capital. However, the increased transaction values also amplify the financial risk should an agent transgress its authority or make an error.

Moreover, B2B payments often involve multiple payment methods such as ACH, wires, instant payments, and cards. An autonomous purchasing agent might eventually need to choose among these methods based on factors like cost, speed, liquidity, or supplier preference. Thus, agents could make microtransactions for AI model queries, price feeds, and API calls before completing a purchase.

Mastercard is currently developing Agent Pay for Machines, which addresses high-speed machine-to-machine transactions. This development transforms payment orchestration into an AI decision problem.

The unresolved issue is determining who should manage the authority accompanying the payment. Banks might anchor this authority to customer identity and account permissions. Payment networks could incorporate it into tokens and authorization standards. Alternatively, AI platforms or merchants could devise their own versions. Waller noted that market participants are already working on both platform-specific and interoperable standards for registering agents, recording approvals, and transferring credentials across commerce systems.