Federal Reserve Identifies Key Barriers to AI-Driven Payments
During his address at Sibos 2026 in Miami on September 29, Federal Reserve Governor Christopher J. Waller provided a detailed analysis of the obstacles that must be overcome for autonomous software agents to handle payments on behalf of consumers and businesses. Waller's speech, titled "Payments in the Age of AI Agents," categorized agentic commerce into two models: agent-assisted, where the buyer retains control, and agent-delegated, which allows the AI to make purchases autonomously within set parameters. The latter model, while promising increased transaction speed, presents significant structural challenges.
Waller identified three primary barriers to the widespread adoption of agent-delegated payments. The first is authentication; the focus shifts from verifying the buyer's authority to ensuring that the agent has the necessary permissions to act on the buyer's behalf. The second barrier is liability; it raises questions about accountability when an agent makes an erroneous purchase. Existing e-commerce frameworks may need to adapt, but Waller suggested that new approaches could also emerge. The third challenge is fraud; current systems designed for human behavior must be recalibrated to accommodate the rapid, machine-driven transactions of autonomous agents.
Market Reading & Resources
Sponsor funding complete. Trade on.
Waller characterized trust as the most significant barrier to scaling agentic commerce, particularly in the agent-delegated model. This perspective aligns with the structural issues that payment networks are already addressing. He framed these challenges as primarily market-structure problems rather than payment-infrastructure issues, indicating that the industry, rather than the central bank, is expected to establish the necessary authentication and liability frameworks.
He cited Visa Intelligent Commerce and Mastercard Agent Pay as examples of card networks that are already developing technical specifications for agent registration and cardholder approval logging. The timing of these developments is notable; for instance, Mastercard expanded its Agent Pay service the day after Waller's speech, introducing probability scoring for transactions initiated by agents and establishing trust signals.
While consumer applications are often highlighted, Waller pointed out that B2B transactions may be more conducive to agentic commerce due to their structured nature, which includes recurring purchases and budget limits. However, the higher transaction values in B2B contexts increase the potential financial risks associated with agent errors.
The complexity of micropayments also adds to the challenges. Agents will need to manage payments for various services, such as LLM queries and API calls, which could favor payment systems with lower fees—a cost structure that remains unresolved in the industry.
Waller also raised the question of whether agentic commerce will operate on interoperable standards across different e-commerce systems or if merchants will create closed ecosystems that limit the use of certain AI agents. He acknowledged that the outcome is uncertain but could significantly impact market structure.
The issues surrounding liability, authentication, and trust have now been recognized at the highest policy levels, with the Federal Reserve mapping out the landscape without prescribing specific standards. The responsibility for addressing these challenges lies with the industry, which is expected to innovate and adapt to the evolving landscape of AI in commerce.




