As Agentic AI Usage Skyrockets, Retailers Face New Challenges and Risks

Artificial intelligence-driven traffic to U.S. retail websites increased 4,700 percent in 2025. Retailers that have developed an agentic AI digital commerce space, or are exploring that possibility, are facing new challenges and risks as consumer use explodes.

What is Agentic AI?

Retailers have to meet current customer expectations by providing an omnichannel shopping experience that is personalized, fast, secure, and as "frictionless" as possible. "Agentic AI" refers to AI systems that can make autonomous, independent decisions and actions to achieve specific human user goals (e.g., "find these Nike sneaker variants up to $250 in price, purchase them, and arrange for home delivery").

Personal electronic assistants aren't necessarily new. Some, such as Amazon's Alexa, have been offering narrow capabilities for specific tasks for more than a decade. However, many of the companies currently building iterative versions of AI technology have commented that AI agents that can carry out complex activities and tasks could be the "killer app" of AI. Tech and payments leaders are already betting on the shift to AI-driven digital commerce, and a growing wave of AI startups is also emerging, with a combination of the two developing the building blocks for fully autonomous shopping.

Who's Really Clicking 'Accept'?

Any retailer familiar with the current state of digital commerce knows the landscape of federal and state regulations and statutes, case law rulings, and payment network rules that set the framework under which a merchant must prove the purchaser's intent and authorization to make a transaction. Where agentic AI adds a wrinkle to the current framework is as follows:

Current State

Under current checkout and payment flows, the human/company making the purchase is involved in both the point of intent ("I want to buy this") and the point of checkout ("I authorize the purchase with my credit card").

Future State Under Agentic AI

Under agentic AI checkout and payment flows, the point of intent and the point of checkout are separated for the first time.

• The point of intent stays with the human who is delegating to the AI agent, and any related merchant terms and conditions likely need to stay with the human at the point of intent level to be enforceable. There should never be "autonomous code" acting solely as "buyer"; rather, the authorization point should be moved up the transaction chain to where the human authorizes the AI agent to take certain actions on the human's behalf within a set of delegated parameters.
• The point of checkout is being delegated by the human to the AI agent under a set of parameters.

But the truly open question and unique issue for agentic AI transactions is who is liable when the AI agent itself malfunctions, such as hallucinating a transaction the human user did not authorize or exceeding the boundaries of the authority delegated to it (e.g., buying 25 pairs of sneakers instead of 2 as instructed by the human user). The company developing the AI agent may try to disclaim all liability, along with direct and indirect damages in its terms of use. But if that's allowed, who gets stuck with the erroneous transaction loss "hot potato" — the user, the merchant, or the issuing bank for the payment method? Retailers need to understand this liability scenario with regard to any proposed agentic AI framework the retailer seeks to adopt.

Card Network and Payment Processing Issues

Developing end-to-end autonomous AI agents for use in digital commerce requires payment authorization processes. Retailers also must pay attention to the various standards emerging from payment networks (and any future standards). Each current approach below places different emphasis on identity, intent, payment control, and standard setting:

• Visa Trusted Agent Protocol (TAP): Visa is emphasizing identify verification by verifying the "who" behind the AI agent. Visa's TAP is tied to Visa's card network and seeks to cryptographically verify in real time that an AI agent making a purchase is indeed legitimate and truly acting on the purchaser's behalf.
• Mastercard Agent Pay: Mastercard is emphasizing tokenization, restricting the "how" of the agentic AI transaction. Mastercard Agent Pay builds on Mastercard's existing tokenization capabilities, creating "Mastercard Agentic Tokens." Mastercard is also partnering with Microsoft Azure OpenAI Service and Copilot Studio to establish pathways for AI systems to complete purchases within conversational interfaces.
• Google Agent Payment Protocol (AP2): Google is emphasizing intent mandates by being able to cryptographically prove the "what" and "why." AP2 is an open, payment-agnostic standard for agents to transact via cards, bank transfers, or even stablecoins and cryptocurrency, using cryptographic user mandates to prove consent.
• Stripe and OpenAI Agentic Commerce Protocol (ACP): Stripe and OpenAI are emphasizing standardized discovery and structuring the "where" to reduce friction and ambiguity by using standard setting and discoverability. ACP is an open-source solution focused on "conversational" checkout and seamless purchase, and utilizes shared payment tokens for AI-mediated transactions in chats/apps.

Emerging Agentic AI Fraud

Finally, retailers need to be aware of (and discuss with their agentic AI partners) how to mitigate emerging fraud attack vectors in the agentic AI space. Some key questions that retailers should be asking include:

• How does agentic AI fraud differ from traditional programmatic fraud attacks?
• How does your service/platform distinguish a legitimate buying agent from a high-speed fraud bot?
• If an AI agent hallucinates and orders 5,000 units instead of 50, who is liable?
• When an autonomous AI agent makes a purchase, who owns the risk (human user, AI agent developer, or the merchant)?

As agentic AI commerce continues its hockey stick growth into 2026 by consumer and business users, retailers that have an existing agentic AI commerce space or are contemplating launching one should think through the agentic AI purchase process flow, partnerships, and payment processing requirements. Issues of unique and emerging risks, payment network requirements, and allocation of transaction liability have to be understood at the front end and baked into the retailer's agentic AI process to ensure sustainability and scalability while guarding against fraudulent use of the agentic AI channel.

Hunton will continue to be a resource and advise our retail clients regarding requirements, risk, and strategies in the agentic AI commerce space.

Erin Fonte is a partner at Hunton Andrews Kurth LLP. She is co-chair of the firm’s financial institutions corporate and regulatory practice.