• NatWest, Bank of America and other banks warn agentic commerce could increase fraud and privacy risks
• Banks want clearer disclosure, data safeguards and transparency around AI purchasing decisions
• Existing fraud systems may struggle to distinguish legitimate AI agents from malicious bots
• Mastercard and Visa are developing controls specifically for agent-initiated payments
• Santander and Mastercard have already completed a live AI-agent payment within a regulated banking framework

Banks are warning that the rapid emergence of artificial intelligence shopping agents could expose consumers to increased fraud, scams and data breaches as technology companies push toward a future in which bots can spend customers' money autonomously.

NatWest, Bank of America, ING, Capital One, Commonwealth Bank of Australia and ASB Bank are among institutions calling for stronger safeguards as agentic commerce moves beyond product recommendations toward AI systems capable of selecting goods and completing transactions.

The intervention comes as OpenAI, Anthropic, Google and Meta develop shopping capabilities for their AI platforms, while retailers increasingly compete to influence the products recommended by chatbots.

Banks said the technology could create risks ranging from incorrect purchases to agents mishandling card information or directing consumers toward payment methods offering weaker protections.

They have called for regulatory discussions covering data safeguards, transparency over how agents make decisions and clear disclosure when AI participates in transactions.

The concerns reflect a broader problem confronting payments risk teams - determining whether existing fraud controls remain effective when the legitimate customer making a purchase may increasingly be a machine acting under delegated authority.

Payments companies are already attempting to address that challenge.

Mastercard has warned that conventional fraud systems could interpret legitimate agent activity as suspicious because bots may transact at unusual hours, across different geographies or make multiple purchases at machine speed.

The company has developed Agentic Tokens intended to identify AI-initiated transactions and link agents to individual users, while its Verifiable Intent technology is designed to provide an auditable record demonstrating that an agent acted within a consumer's instructions.

Visa is taking a similar approach. Its Intelligent Commerce infrastructure embeds payment credentials, authentication and controls into AI-initiated transactions.

The company has also partnered with OpenAI to support payments within agentic commerce.

Consumer appetite, however, appears to be running ahead of willingness to surrender financial control completely.

Visa said this month that 72% of US consumers have used an AI assistant alongside search engines to discover products, but acknowledged that allowing an agent to actually make the purchase represents a substantially greater trust hurdle.

The industry's concerns are no longer theoretical. Santander and Mastercard completed what they described as Europe's first live end-to-end payment executed by an AI agent within a regulated banking framework in March.

The transaction used predefined limits and permissions intended to ensure that the agent could act only within parameters established by the customer.

The question of accountability becomes more complicated as autonomous systems gain greater freedom.

If an agent purchases the wrong product, falls victim to a fraudulent merchant or makes a transaction outside what the customer intended, banks, card networks, technology providers and merchants may face difficult questions over who carries responsibility and whether established consumer protections apply.

The banks calling for safeguards have consequently emphasized the importance of transparency and consumer choice, including the ability of customers and merchants to choose AI services and for different platforms to operate together.

Agentic commerce promises to remove considerable friction from online shopping. But payments providers have spent decades building systems around identifying customers, verifying intent and detecting unusual behavior.

AI agents potentially change all three simultaneously - leaving risk teams with the challenge of ensuring that making payments easier for machines does not inadvertently make fraud easier too.