Six Banks Warn of Fraud Disputes from Agent Shopping
On Tuesday, Bank of America, Capital One, NatWest, ING, New Zealand's ASB, and Australia's Commonwealth Bank jointly released a principles document titled "Building Trust in Agent Commerce," stating that granting shopping chatbots greater autonomy may increase rates of fraud, scams, and disputes, with risks covering five categories: transparency, security, privacy data, choice, and interoperability.
The document specifies that consumers are unclear whether agents are on their side, worrying about making wrong purchases or overspending, and are more concerned about not knowing whom to seek compensation from after being defrauded. Specific pathways include agents requesting card numbers and directly entering them on websites, or directing payments to less secure channels; attackers can impersonate or hijack agents and merchants, using new social engineering tactics.
Issuing banks and acquiring banks often cannot see the identity of the agent, the merchant entity, the user's original intent, and purchase details, making it difficult to allocate responsibility along the value chain when issues arise. Banks demand to retain auditable records from instructions, authorizations, intents to payment results, and to include all relevant parties in the dispute process. Mark Monaco, head of global payments at Bank of America, emphasized identity, authorization, fraud, and liability management; Todd Kennedy, head of payment strategy at Capital One, stated the need to shape a secure ecosystem from the customer side.
The six institutions plan to discuss with policymakers: transactions must disclose whether they were completed by an AI agent, the decision-making process must be explainable, and customer data must be separately protected. The document also acknowledges customer enthusiasm for agent commerce, stating it could become a mainstream transaction method, but standards lag behind technology. Further implementation details will be released later.
ChatGPT's instant checkout has covered over 1 million Shopify stores and Etsy, but still requires user confirmation for orders and payments. UK retailer John Lewis reported that search traffic from AI agents rose from 0.3% to 2.5% within a year. Another survey indicated that over two-fifths of Americans have encountered or nearly encountered AI scams.
Those buying are banks wanting to write card organization rules into the agent layer, while those selling are model companies competing for checkout entry. Funds are shifting from merchant cash registers to in-dialog payments, with disputes and chargeback costs initially hitting issuing banks and merchants. Beneficiaries are banks and card networks that still hold the right to refuse payments, while those under pressure are agent products that explicitly fill in card numbers on third-party pages and platforms unable to prove "user intent."
The six banks clearly list commission and token cost-driven goods and payment recommendations as conflicts of interest, requiring agents not to prioritize user interests for higher splits.
Source: Public Information
ABAB AI Insight
Issuing banks profit from chargebacks and liability allocation while avoiding risks. Bank of America and Capital One frame payment innovation as customer experience, NatWest and ING bring cross-border card organization rules into Europe, while the two federal banks in Australia and New Zealand apply the same principles to another time zone. The joint publication is not a technical paper, but an industry memorandum written before ChatGPT's checkout expansion, addressing "who bears the loss."
The capital pathway is to keep card numbers within a bank-controlled track. Once agents fill in cards for users or select weaker payment methods, banks lose visibility over tokenization and 3DS layers, yet chargebacks still fall on issuing banks. Thus, the core of the five principles is the audit trail: instructions, authorizations, intents, and results must be traceable; otherwise, the Visa/Mastercard liability framework cannot apply to conversational transactions. Banks need to negotiate mandatory disclosures with regulators, effectively trading compliance costs for payment routing rights.
This is analogous to the mobile wallet wars from 2014 to 2016: Apple Pay replaced card numbers with tokens, banks initially resisted but later invested in the rules; PayPal checkout also faced high dispute rates due to unclear liability. Today, agent commerce is in the "expanding transactions, yet to define responsibility" phase, with OpenAI and Shopify embedding checkouts into chats, while banks redefine checkout as regulated payments. John Lewis's agent traffic rising from 0.3% to 2.5% indicates volume is increasing, but rules remain on paper.
Structural changes are driven by regulatory shifts and the transfer of pricing power. Pricing power is shifting from merchant page buttons to model recommendations and payment method choices; regulatory gaps allow scammers to impersonate agents, while banks can demand a seat at the negotiation table under the guise of consumer protection. The mechanism is: as autonomy rises, the burden of proof shifts from user screenshots to full-chain logs; whoever enforces logs first will continue to be the final settlement party.
ABAB News · Cognitive Laws
As autonomy rises, the responsibility vacuum expands.
The checkout entry determines who loses money more than the recommendation algorithm.
When standards lag behind technology, banks use principle documents to seize legislative seats.