Amazon and Walmart AI Shopping Bots Face Calls for FTC Scrutiny
The two developments sit next to each other in an AI-shopping workflow, but they are not one transaction or one control failure. Retailers need to preserve material product facts; payment networks need to constrain delegated spend.
U.S. Senators Tammy Baldwin and Rick Scott asked the Federal Trade Commission to examine how AI shopping assistants at Amazon and Walmart present "Made in USA" information, according to the Wall Street Journal. The request cites research alleging that assistants can fail to surface or correctly interpret product-origin information. The FTC has not made a finding of wrongdoing.
Amazon told the Journal that any suppression of origin information is unintentional and that it is improving the capability. The immediate issue reaches beyond one label. A conversational shopping agent compresses dozens of product pages into a small set of recommendations, which means the system decides which seller, origin and product claims survive that compression.
Mastercard and Alchemy are working on a different problem. Their agent-payment tools use virtual or tokenized credentials that can be constrained by amount, merchant and other user-defined rules. Mastercard's broader Agent Pay architecture is built around delegated authority rather than handing software an unrestricted reusable card credential.
There is no evidence in the material reviewed that the Amazon or Walmart assistants cited by the senators use Mastercard's or Alchemy's payment architecture. The connection is the workflow, not a shared deployment: recommendation integrity sits upstream of payment authorization, and either layer can fail while the other works as designed.
For a retailer, the audit record has to show which product facts the recommendation system saw and whether material disclosures survived summarization. For a payment network, it has to show what the user authorized the agent to buy, from whom and within what spending limit. A provenance failure is not repaired by a valid payment credential, and a correct recommendation does not authorize an agent to spend freely.
The near-term commercial burden is split. Retailers need better structured product data and recommendation provenance; payment networks need constrained credentials, identity and transaction logs. Treating those as one market would overstate the connection between two adjacent control layers.