News·5 min read

Anthropic's AI Shopping Agents Add a New Trust Layer on Payments

Anthropic launched a retailer blueprint for Claude shopping and merchant agents, with Visa and Mastercard support. The AI can build the basket; payment still sits on the merchant's checkout - until that last step changes.

IH

By

Co-Founder & CEO

Anthropic has launched a blueprint for retailers to build AI shopping and merchant agents with Claude, with support from Visa and Mastercard. For now the agent can search, recommend, and build the basket, but the final payment still goes through the merchant's existing checkout. When an AI can transact on someone's behalf, fraud questions go beyond "is this person authorised to use this card" to whether the agent, the mandate, and the purchase are authorised too.

Anthropic has launched AI shopping agents with support from Visa and Mastercard. They launched a blueprint for retailers to build AI-powered shopping and merchant agents using Claude. This is one of the biggest developments in agentic commerce.

The shopping agent can search a retailer's catalogue, remember customer preferences, answer questions, recommend products, and build the customer's basket.

There's an important distinction. For now, the AI can take the customer right up to the point of payment, but the final transaction still happens through the merchant's existing checkout or payment provider.

From a fraud perspective, I think what happens when that final step changes is where things get particularly interesting. Once an AI agent can transact on someone's behalf, determining whether a transaction is legitimate becomes more complicated.

Today, we're largely trying to establish whether the person making a payment is who they claim to be and whether they're authorised to use that payment method. With agentic commerce, another layer gets introduced. Now we potentially need to establish:

  • Is this a legitimate AI agent?
  • Has the customer authorised this agent to transact?
  • Does this specific purchase fall within that authorisation?
  • Are the payment credentials being presented securely?
  • What happens if that authority is revoked?
  • Who is responsible if the agent makes the wrong purchase?

This is why I think Visa and Mastercard's involvement is particularly significant. AI might increasingly control the interface where products are discovered, compared, and selected, but payments infrastructure will still need to determine whether the resulting transaction is authorised, secure, and accountable.

Agentic commerce is adding an entirely new trust layer on the existing payments stack. As AI moves from recommending what we buy to buying it for us, that trust layer is going to become increasingly important.

Related: if checkout pages go away, fraud systems have to change what they ask, how it works, and the FAQ.

If you want ranked rule changes on the stack you already run - Stripe Radar, Shopify Flow, Blockify, or Adyen RevenueProtect - book a demo.

Originally shared on LinkedIn.

FAQ

Frequently asked questions

More buyer questions on Radar, Flow, Blockify, chargebacks, and Signifyd alternatives.

View FAQ

Ready to see it
on your data?

Book a live walkthrough and see how FraudPulse turns your payment data into actionable fraud intelligence.