Imagine asking an AI assistant to find a product, compare prices and buy it without visiting multiple websites. That experience is becoming possible as artificial intelligence (AI) moves beyond recommendations and enters online checkout.
AI Shopping agents can now handle several steps of the buying process on behalf of consumers. Instead of searching different websites, comparing prices and entering payment details, shoppers can delegate these tasks.
OpenAI has taken this further with ChatGPT Wallet, introduced on November 13, 2026. The feature allows its AI agents, called dots, to make purchases using customers’ cards without accessing actual card details.
But would you trust an AI assistant to spend your money without checking every purchase? What happens if it orders the wrong product, exceeds your budget or encounters a security breach?
These questions are attracting attention from banks and payment companies. Their concerns go beyond payment security to include fraud, consumer protection and responsibility for purchases made by autonomous AI agents.
AI Shopping Agents Are Moving Beyond Product Recommendations
Traditional shopping assistants help consumers find products, compare prices and check customer reviews. However, shoppers still choose what to buy and complete the payment themselves.
AI shopping agents are changing that process by taking part in online purchases. Consumers can instruct these agents to search for products, compare options and interact with retailers during checkout.
In a September 29, 2026, speech, Federal Reserve Governor Christopher Waller outlined two approaches to agentic commerce: agent-assisted shopping and agent-delegated shopping. Agent-assisted shopping allows AI systems to research products, compare prices and recommend suitable options. However, consumers retain control over purchasing decisions and payments.
Agent-delegated shopping gives AI agents permission to select products and complete purchases independently. Consumers can set spending limits and other conditions that determine what an agent can buy.
How AI Shopping Agents Use Your Money to Make Purchases
AI shopping agents do not always need direct access to bank accounts or credit card details. Payment systems can use tokenization, allowing agents to initiate purchases without exposing sensitive financial information. When consumers authorize an AI agent to shop, the payment process typically involves several steps:
- Payment Authorization: Consumers connect an eligible payment method and grant the AI agent permission to make purchases.
- Shopping Instructions: Consumers specify the products they need, preferred price ranges and spending limits.
- Product Selection: The agent searches online retailers, compares prices and selects products matching the consumer’s instructions.
- Payment Processing: The agent uses an authorized payment token to initiate checkout without accessing original card details.
- Transaction Verification: The payment system performs security checks before approving and processing the transaction.
- Order Confirmation: The retailer receives the order, while the payment provider processes the financial transaction.
Payment Tokenization helps protect sensitive payment information throughout this process. It replaces original card details with unique digital tokens that support authorized transactions.
ALSO READ: Meta’s New Muse AI Agent Can Send Emails, Book Travel and Make Payments
OpenAI’s ChatGPT Wallet Brings AI Shopping Payments to Consumers
OpenAI introduced ChatGPT Wallet on November 13, allowing its dots agents to complete purchases on users’ behalf. The payment feature was developed with American Express, Chase, Mastercard and Visa. It uses payment tokens to protect customers’ actual card details during transactions.
Users must still approve purchases before ChatGPT completes any payment. The checkout process displays selected products, the total amount and the payment card. Customers then verify transactions through security checks, including Face ID or fingerprint authentication. After approval, the payment network generates a token that replaces actual card information.
This process uses existing payment infrastructure without exposing sensitive card details to AI agents or retailers. OpenAI says eligible purchases retain existing card rewards and protections for merchant disputes. Retailers remain responsible for product pricing, delivery, returns and customer service.
The initial rollout targets ChatGPT Pro subscribers aged 18 and older across the United States. Users also need eligible US-issued payment cards and a US billing address. OpenAI plans to introduce more autonomous purchasing capabilities in future updates.
These options could let customers authorize purchases within predefined budgets, product categories and time limits. However, these capabilities remain part of OpenAI’s longer-term plans. ChatGPT Wallet currently requires customers to approve every transaction before payment.
What Happens If an AI Shopping Agent Spends Your Money Incorrectly?
The biggest risk with AI shopping agents is their ability to make mistakes involving consumers’ money. An agent could misunderstand instructions, select unsuitable products or overlook specifications that matter to the buyer. It could also choose expensive alternatives that exceed the customer’s intended spending limit.
Such errors could lead to unwanted purchases, duplicate orders or products that fail to meet expectations. Security presents another serious concern for consumers using AI shopping agents. Criminals could exploit compromised accounts or manipulate agents into making fraudulent purchases.
Financial institutions have already started raising concerns about these potential threats. A September 22 Reuters report highlighted warnings from major banks, including NatWest, Bank of America, ING and Capital One.
These institutions warned that AI shopping agents could expose consumers to more scams, financial fraud and privacy breaches. They also questioned whether agents might steer shoppers towards payment methods offering weaker consumer protections.
ALSO READ: Companies Are Building AI Clones of Consumers to Predict What You’ll Buy
The warnings show a growing gap between agentic commerce development and the safeguards needed to protect consumers. Protecting payment card details alone will not address every risk associated with autonomous shopping. Financial institutions must also verify whether agents have proper authorization before approving individual purchases.
Who Is Responsible When an AI Agent Makes the Wrong Purchase?
Financial responsibility remains a major unanswered question as AI agents gain permission to make purchases. If an agent buys the wrong product or exceeds spending limits, several parties could face disputes. These parties may include the AI provider, payment network, issuing bank and retailer.
Their responsibilities will depend on the transaction details and applicable consumer protection rules. Federal Reserve Governor Christopher Waller identified authentication, liability and fraud as major challenges facing agentic payments. He said payment systems must verify whether agents have permission to spend money on consumers’ behalf.
This creates a new challenge for financial institutions handling automated payments. Traditional systems authenticate the person making payments, while agentic commerce must also verify delegated permissions.
The Bank for International Settlements raised similar concerns in its November 9, 2026, speech. It said consumers must understand agent instructions, monitor their actions and challenge transactions.
Consumers must also retain the ability to revoke permissions granted to AI agents. Consider someone authorizing an AI agent to purchase groceries within a $100 spending limit. If the agent spends $180, questions arise about whether the transaction exceeded its authorization.
The situation becomes more complicated when agents stay within budget but purchase incorrect products. Existing card protections may cover certain merchant disputes, but they do not resolve every agent-related purchasing error.
OpenAI says ChatGPT Wallet retains existing card rewards and protections for eligible merchant disputes. However, it has not confirmed reimbursement for every mistake involving AI shopping agents. Clearer industry rules will be necessary to determine financial responsibility when automated purchases go wrong.
How Consumers Can Protect Their Money When Using AI Shopping Agents
AI shopping agents can simplify online purchases, but consumers should set clear financial limits before granting purchasing permissions. Several precautions can help prevent unwanted purchases and reduce financial risks:
- Set spending limits: Limit how much agents can spend per purchase or within a specific period.
- Use virtual payment cards: Choose tokenized payment options instead of sharing actual card details with shopping agents.
- Require purchase approval: Check products, sellers, delivery information and final prices before approving transactions.
- Restrict purchasing categories: Allow routine household purchases while blocking expensive or sensitive product categories.
- Monitor transactions: Review bank statements and payment alerts regularly to identify unfamiliar or unauthorized charges.
- Check return policies: Confirm whether retailers accept returns and offer refunds before authorizing purchases.
Consumers should also know how to revoke an agent’s purchasing permissions. This becomes important if accounts are compromised or agents repeatedly make incorrect purchases.
OpenAI’s ChatGPT Wallet already requires transaction approval and uses payment tokens to protect card information. Its future plans include spending limits and category restrictions for more autonomous purchasing.
Conclusion: AI Shopping Needs Stronger Consumer Protection Before It Scales
The future of AI agentic commerce depends on whether consumers trust AI agents to handle their money responsibly. Technology companies are building payment systems that protect card details during automated transactions.
However, secure payments cannot prevent incorrect purchases or resolve every financial dispute. Banks, retailers and AI companies need clearer rules covering authorization, refunds and financial liability. They must also give consumers ways to monitor purchases and revoke permissions.
Fraud prevention presents another challenge as AI agents become more common. Existing detection systems may need updates to recognize transactions initiated by software rather than humans.
ChatGPT Wallet’s limited US rollout shows that AI shopping is entering real-world payment systems. However, fully autonomous purchasing remains a developing capability rather than today’s standard experience.
AI agents could eventually become regular participants in online shopping. But allowing software to spend independently requires stronger protections than simply recommending products.

