Target Rewrites Its Terms So Customers Eat the Cost When the Gemini Shopping Agent Buys the Wrong Thing
Target updated its consumer terms to treat any purchase placed by an Agentic Commerce Agent as a transaction "authorized by you," shifting AI mistake liability to shoppers ahead of its Gemini-powered checkout launch.
Overview
Target has quietly become one of the first major U.S. retailers to put in writing what most of the agentic commerce industry has so far left ambiguous: when an AI shopping agent acting on a customer’s behalf buys the wrong thing, the customer pays. The change, contained in an update to the company’s online consumer terms, is timed to Target’s pending integration with Google’s Gemini assistant and the Universal Commerce Protocol that lets shoppers complete purchases without leaving a chat window.
What Target Changed
Target’s revised terms introduce a new section governing what the company calls an “Agentic Commerce Agent” — a third-party AI tool that a shopper authorizes to interact with Target on their behalf. Under the updated language, transactions completed by such an agent are “considered transactions authorized by you,” regardless of whether the outcome matches the user’s actual intent. The same clause states that Target “does not purport to guarantee that an Agentic Commerce Agent will act exactly as you intend in all circumstances” and places the duty to monitor agent activity squarely on the shopper.
In practical terms, the policy means that if a Gemini-powered agent ordering on a shopper’s behalf buys the wrong size, the wrong variant, or a more expensive version of a product than the customer wanted, the resulting charge is enforceable against the customer. Reached for comment, a Target spokesperson told Newsweek that “new AI tools make shopping at Target even easier by recommending items and helping complete purchases at the guest’s direction” and emphasized that purchases routed through the agent remain eligible for the company’s standard return policy. That carve-out preserves the existing refund pathway but does not undo the underlying liability assignment for the original transaction.
Newsweek also reported that Target is partnering with OpenAI on the assistant’s underlying intelligence in addition to the Gemini integration, with OpenAI Applications CEO Fidji Simo quoted as saying the company is “excited to work with Target as they weave intelligence throughout their business to create useful and joyful experiences.”
Why Target Wrote This Now
The terms-of-service revision is the consumer-facing legal scaffolding for a checkout integration that Google announced at the National Retail Federation conference on January 11. In that announcement, Google introduced the Universal Commerce Protocol — an open standard “for agents and systems to operate together across consumer surfaces, businesses and payment providers” — and named Target as a co-developer alongside Shopify, Etsy, Wayfair, and Walmart, with more than 20 additional endorsers including American Express, Mastercard, Stripe, Visa, Best Buy, Macy’s, and The Home Depot.
The protocol is designed to let AI agents handle the entire shopping lifecycle — discovery, checkout, and post-purchase support — without bespoke per-retailer integrations, and it interoperates with the Agent2Agent, Agent Payments Protocol, and Model Context Protocol standards that have emerged over the past year. Google said UCP would soon power a new checkout feature in AI Mode in Search and the Gemini app, with Google Pay handling payments and PayPal support to follow. The Machine Herald previously reported on Google’s March expansion of UCP to add Cart, Catalog, and Identity Linking primitives — the technical building blocks that let an agent assemble a multi-item order and complete checkout without bouncing the user back to a retailer’s website.
The pieces are now in place: the protocol exists, the agent platform exists, and the contract that governs what happens when something goes wrong now exists too.
The Liability Vacuum Target Is Filling
U.S. consumer-protection law has not yet been adapted to autonomous AI agents that complete transactions, and no federal agency has issued binding guidance on who bears the cost when an agent makes a purchase the human did not specifically intend. The Federal Trade Commission has issued general statements about AI and consumer protection but has not addressed agentic commerce specifically. Existing payment-protection rules — the Electronic Fund Transfer Act for debit cards and the Fair Credit Billing Act for credit cards — determine when a transaction is “unauthorized” for the purposes of chargeback rights, but those statutes were written long before software agents could place orders on a consumer’s behalf, and the courts have not yet tested how they apply to a fact pattern in which the consumer voluntarily granted authority to an AI.
Target’s contract is an attempt to settle the question on Target’s preferred terms before regulators or judges reach a different one. By labeling the agent’s purchases as “authorized by you,” the retailer is trying to slot agentic commerce into the existing framework that treats card-not-present transactions placed with a customer’s credentials as the customer’s responsibility, rather than into the framework that treats automated, unintended charges as unauthorized.
Industry coverage suggests other retailers are watching closely. SiliconAngle described UCP as a system designed to let AI agents automate “the entire shopping journey, from product discovery to payment to post-sale support” across a roster of major merchants, and TechCrunch noted that Google positioned the protocol as a way for retailers to participate in agent-driven shopping without separate per-agent integrations. Whether the rest of the UCP coalition follows Target’s lead in formally codifying customer liability for agent errors is likely to determine how quickly the industry’s terms of service converge on a single answer.
What We Don’t Know
The exact public launch date for Target’s Gemini-powered checkout has not been disclosed. It is also unclear how Target will handle disputes that fall between an unambiguous AI error and a clear customer instruction — for example, an agent that orders a slightly different SKU than the one a shopper described in natural language. The standard return policy provides recourse for unwanted items, but returns require the customer to identify the problem, ship the product back, and absorb any non-refundable fees in the meantime.
It remains unsettled whether existing federal payment-protection rules will treat an AI-agent purchase as “authorized” in the same way Target’s contract does. Card networks, regulators, and the courts have not yet tested that question against an agentic commerce fact pattern, leaving a gap between merchant terms of service and the underlying payment law that ultimately governs whether a charge sticks. Until that gap closes, the practical answer for shoppers will be whatever the retailer’s contract says — and at Target, that contract now says the bot is the buyer’s responsibility.