Can AI Agents Buy Things Without Your Permission?

Yes, an AI agent can buy things without asking you at the moment of purchase if it has been given access to a payment method and is allowed to complete transactions. That does not mean every agent can spend freely: its ability depends on the tools, permissions, spending limits, and confirmation steps configured for it.

What an AI agent can actually do

An AI agent is software that can use tools to pursue a goal, such as searching websites, comparing options, filling in forms, or calling a shopping service. Unlike a chatbot that only recommends a product, an agent connected to a checkout flow may be able to place an order. The crucial distinction is not whether the system is called “AI,” but what accounts and actions it can access.

For example, an agent might be authorized to find a replacement printer cartridge and prepare a purchase. If it can submit the order using a saved card, and the service does not require an additional confirmation, it may complete the transaction without asking you again. Another setup might let it browse and fill the cart but stop before the final purchase button.

“Without permission” can mean two different things:

Without a fresh confirmation: You previously authorized the agent or connected service to make purchases under certain conditions.
Without any valid authorization: The agent acts outside the permissions you intended, perhaps because access was misconfigured, compromised, or exploited.

The first can be an expected feature. The second is a security or accountability problem.

How permission and checkout work

An agent usually does not independently possess money. It acts through access granted by a user, an organization, or a connected service. That access may include a shopping account, a digital wallet, a payment token, or an API that can create orders. Whether payment completes depends on how those systems handle authentication and approval.

A typical purchase flow may involve several separate permissions:

Finding products: The agent can search catalogs or read product details.
Preparing an order: It can select an item, quantity, shipping address, or delivery option.
Submitting the order: It can send the checkout request to the seller.
Authorizing payment: The payment provider or account may require a password, device approval, one-time code, or other check—or may rely on an existing authorization.

These steps do not always happen in one place. A shopping app may permit the agent to prepare a cart while the payment provider independently applies its own authentication rules. A saved card may make checkout convenient, but it does not automatically mean every agent has permission to use it. The actual behavior is determined by the service settings, credentials, and safeguards in that particular setup.

Why an agent might act unexpectedly

Agents can misunderstand vague instructions, choose a poor match, or interpret a recommendation as permission to buy. They can also be influenced by misleading information in a webpage, email, or product listing. A system that reads external content and can take actions must distinguish instructions from the user from text that merely appears in material it is processing; that distinction can be difficult to enforce perfectly.

Other causes are more ordinary: a broad authorization, an unnoticed recurring order, a duplicate request after a slow response, or a connected account that remains signed in. An agent may also select a different size, shipping method, or seller than expected if the instruction did not specify those details. These are reasons to make the agent’s allowed actions explicit rather than assuming it will infer your boundaries.

How to reduce the risk

Before connecting an agent to a shopping or payment account, check what it can do and how to revoke access. Prefer a setup that separates research from purchase approval, especially when an order is costly, recurring, or difficult to cancel.

Require confirmation before an order is submitted or payment is authorized.
Set a spending cap, allowed item categories, or approved sellers if the service supports those controls.
Limit connected accounts and permissions to what the task needs; do not share a main account password when a narrower authorization option is available.
Review the item, quantity, total, seller, delivery address, and renewal terms before approving.
Enable transaction notifications and check order history and payment statements regularly.
Remove the agent’s access when you stop using it, and secure the account through the service’s normal account-recovery and authentication settings.

For business use, organizations should define who may authorize purchases, set approval thresholds, and keep records of requests and completed transactions. A human review step is particularly useful when the agent can spend from shared funds or place repeat orders.

If an unwanted purchase happens

Act promptly. Check the order in the seller’s account and use the seller’s available cancellation or return process. Contact the payment provider if the transaction appears unauthorized, and follow its instructions for reporting or disputing it. Also revoke or narrow the agent’s account access, change credentials if they may have been exposed, and review other recent activity.

Keep relevant records, such as the agent’s instruction history, order confirmation, and account alerts. These can help establish whether the purchase followed an existing authorization, resulted from an error, or involved access you did not approve. Do not assume that deleting a chat or uninstalling an agent cancels an order or revokes permissions held by a separate connected service.

Frequently Asked Questions

Can ChatGPT or another AI buy something for me?
That depends on the product, available integrations, and permissions you enable. Some systems may only provide recommendations; others may be connected to tools that can prepare or submit orders. Check the specific service’s purchase and confirmation settings.

Can an AI agent use my saved credit card?
Only if the agent or connected service has a way to access a checkout flow authorized to use that payment method. A saved card alone does not show which tools have permission, so review connected apps and account settings.

Will an AI agent always ask before paying?
No. A confirmation prompt depends on the agent’s design and the merchant or payment provider’s rules. Do not assume there will be a last-minute prompt unless the service clearly provides and enforces one.

Can an AI agent make recurring purchases?
It may be able to place repeat orders if it has the required account access or if you enable a subscription or auto-reorder feature. Check renewal terms and disable recurring orders through the seller or service that manages them.

What should I do if an AI agent bought something I did not want?
Check the order and contact the seller promptly about cancellation or return options. If you believe the payment was unauthorized, contact your payment provider, revoke the agent’s access, and review recent account activity.

The practical answer

AI agents can make purchases without a new prompt when their connected tools and account permissions allow it. They should not be treated as automatically safe or automatically unauthorized: the meaningful questions are what actions you granted, whether approval is required at checkout, and how easily you can monitor and revoke access. Give an agent only the authority needed for its task, and keep a clear confirmation step for purchases you would not want it to make on its own.

Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.

More cards in this drawer

The whole Numbers and Powers drawer →