SI Agent News
DevelopingOpenAIAgentsPolicyResearch

OpenAI launches ChatGPT Wallet for US Pro users with biometric checks

The new feature allows AI agents to handle online checkouts using payment tokens, requiring a Face ID or fingerprint for every transaction.

$100monthly cost for ChatGPT Pro

The short version

Why it matters

This marks a shift toward 'assisted checkout' where AI handles the logistics of shopping while humans retain final veto power. It highlights the ongoing tension between AI automation and the security requirements of traditional payment networks.

The launch of ChatGPT Wallet

OpenAI officially introduced ChatGPT Wallet on November 13 to facilitate online shopping through its 'dots' agents. These agents are designed to run on cloud computers with their own browsers, allowing them to connect to more than 4,000 different apps. The wallet is currently rolling out to US-based ChatGPT Pro subscribers who hold eligible cards from networks like Visa, Mastercard, American Express, and Chase.

The feature is specifically designed for use within dots conversations rather than standard ChatGPT chats. Because the rollout is limited, OpenAI is currently working with various banks and payment networks to ensure compatibility. The company intends to expand the service as it gathers data from early users, though no specific timeline has been provided for free users or international customers.

While the wallet allows for a more seamless shopping experience, it does not hold a cash balance or cryptocurrency. Instead, it acts as a bridge between the AI's ability to navigate websites and the user's existing banking information. It aims to capture the final checkout step for the 1.2 billion people who use ChatGPT weekly to research and compare products.

How payment tokenization works

The system relies on a process called payment tokenization to keep sensitive data secure. When a user adds a card, the network issues a payment token, which is a surrogate credential that has no exploitable meaning outside of the specific system. This means the merchant and the AI agent never see the user's actual card number, security code, or expiration date.

For every transaction, the dot agent presents the items, the total cost, and the selected card to the user. Once the user provides a biometric scan, the card network generates a token scoped to that single agent-driven transaction. This setup ensures that even if a merchant's page is compromised, the underlying card details remain protected from theft.

This method differs from other tools like Instant Checkout, which OpenAI previously launched with Stripe. Because the token travels through existing merchant checkout systems, businesses do not need to build a separate integration to accept these payments. However, the token does not currently have a hard amount cap or an expiration date, though some frameworks are beginning to add those features.

Security and human oversight

OpenAI emphasizes that the biometric check is the most important control because the token only secures the credential, not the agent's reasoning. A major concern is 'prompt injection,' where a malicious website might hide instructions to steer the AI toward unwanted purchases. Since there is no surefire fix for these types of attacks, human oversight remains the primary defense for sensitive operations.

The current setup is what researchers call 'assisted checkout' rather than 'true agentic payments' because it requires a human to click 'approve' every time. OpenAI does plan to eventually allow users to authorize agents to buy within preset limits, such as specific spending caps or product categories. However, these automated safeguards are not yet available to the public.

The company is currently working with several fraud-prevention firms and payment providers to refine these systems. They are also exploring WebMCP tools as another way for websites to support agent-driven commerce. For now, the user must treat the review screen as the official record of what they are authorizing the AI to buy.

Disputes and merchant restrictions

One of the biggest unanswered questions is how federal rules apply when an approved agent purchase goes wrong. Chargeflow reported that there is currently no specific guidance from the Consumer Financial Protection Bureau (CFPB) regarding AI agents. Whether a mistake by an agent counts as 'unauthorized use' or a 'billing error' remains a grey area in current regulations.

Credit cardholders generally fall under Regulation Z, while debit cardholders fall under Regulation E, each with different liability caps and reporting windows. If a user deliberately gives an agent the power to buy things, a dispute might be handled as a merchant error rather than fraud. Some card issuers may even view these charges as authorized because the human provided the final biometric sign-off.

Additionally, some major retailers like Amazon and eBay have policies that restrict outside AI agents from placing orders. eBay's user agreement specifically bars bots that place orders without human review unless they are approved by the platform. Users should be aware that even if they approve a purchase in ChatGPT, the merchant may still block or decline the order based on their own fraud filters.

What to watch

Try it

If you are a US ChatGPT Pro subscriber, check your support documentation to see if your specific card from Amex, Chase, Mastercard, or Visa is eligible for the Wallet.

Sources and links

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

Listen to the AI Wire headlines

Listen to the AI Wire

Listen to our daily AI shows every morning: on your commute, in the shower, wherever you are

Listen to the daily shows

First reported by Tech Times · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

More AI news