What blocks autonomous agents from handling purchases at scale?

Christopher J. Waller, in a Federal Reserve Board speech, splits agentic commerce into two models: in the "agent-assisted" model "a buyer uses an AI agent primarily for product search and discovery, but the buyer remains in control. They make the decisions, and they handle the payment." In the "agent-delegated" model, "a buyer grants authority to an AI agent to shop and make payments on their behalf," with the buyer only setting constraints and guardrails while "the agent operates autonomously." Waller states plainly that "the biggest barrier to scaling agentic commerce, particularly the agent-delegated model, is building sufficient trust among buyers and sellers. A few concrete challenges stand out: authentication, liability, and fraud." He ties the elevated risk directly to delegation: "Because it introduces higher risks of unintended purchases, the delegated model requires a more extensive buildout of trust mechanisms and guardrails."

Waller unpacks each challenge in the payment industry's own terms. Authentication: "agentic commerce transforms the payment authentication paradigm. The question shifts from proving that a buyer is an authorized payer to proving that an agent has the authority to pay on the buyer's behalf." Liability: "Who is on the hook if an agent makes the wrong purchase?" He notes existing e-commerce liability frameworks "could potentially be adapted for agentic commerce," while also flagging "an opportunity to test and refine new approaches," including technical standards that record "what the buyer intended and how their agent carried it out" to reduce dispute ambiguity. Fraud: "Existing fraud detection and prevention systems, calibrated to human behavior, may not translate well to agents," so "fraud models and rules will need to be recalibrated to account for agent payment patterns."

This sits alongside Can a protocol split agent access from agent incentives?, which is one concrete attempt at the authority-granting mechanism Waller's agent-delegated model requires — consumers deciding access, companies bounding what agents can do, mapping onto his constraints and guardrails. It also complements What infrastructure must exist beyond training agents?: Waller's authentication challenge is the payments-industry instance of that paper's attribution function, and his fraud challenge is an instance of its response function, arriving from a central banker's practical framing rather than a research taxonomy. And it is the policy-side counterpart to Can existing identity standards work for AI agents?, which proposes the technical identity standards (OAuth, OIDC, SPIFFE/SPIRE) that Waller's authentication problem — proving an agent's authority to pay — would need to draw on.

The excerpt is a speech, not a study: Waller offers no data on actual fraud rates, liability disputes, or authentication failures in agentic commerce, and frames the three challenges as open questions for the industry to solve rather than findings about how often they occur or how costly they are. He explicitly leaves unresolved "what standards are still missing for agents to carry identity, consent, and payment credentials across the full e-commerce stack" and "what is the right balance between user convenience and purposeful friction in agent authorization." The implication the text supports is narrower than these problems are being solved — it is that a sitting Federal Reserve governor is naming authentication, liability, and fraud as the specific bottleneck the industry must clear before agent-delegated commerce scales, without evidence yet that any of the three has been resolved.