Amazon has blocked Meta’s Muse personal AI agent from browsing and purchasing products on Amazon.com, highlighting an emerging challenge for agentic commerce: who controls the customer relationship when an AI agent shops on a consumer’s behalf?
The move comes less than two weeks after Meta launched Muse, an AI agent designed to perform longer-running tasks for users, including shopping.
Amazon Raises Security and Control Concerns
According to Amazon, it had not authorized Muse to access its store, customer accounts, scrape data or process transactions. The company said it was concerned about customer security and the shopping experience and argued that third-party agents should respect a service provider’s decision about whether to participate.
Users attempting to shop on Amazon through Muse began seeing a message stating that continued access by an unauthorized AI agent violated Amazon’s conditions of use.
The Bigger Issue: Who Owns the Shopping Journey?
The dispute points to a fundamental question for agentic commerce.
Traditionally, shoppers visit a retailer’s website, search for products, compare options and complete purchases within the retailer’s environment. With AI agents, much of that journey could happen outside the retailer’s platform.
An agent could potentially:
- Understand what a consumer wants
- Search across multiple retailers
- Compare products and prices
- Select an option
- Add it to a cart
- Request user approval
- Complete the transaction
That could reduce the amount of time consumers spend directly interacting with individual retailer websites.
Axios cited Chris Jones of PSE Consulting as describing the issue around who owns customer origination and the customer relationship versus who simply fulfills the transaction.
Amazon and Meta Take Different Positions
Amazon says third-party applications making purchases for customers should operate openly and respect individual service providers’ decisions.
Meta, meanwhile, has designed Muse to operate through a dedicated virtual machine and has introduced controls around internet access, permissions and user approval for sensitive actions.
The disagreement illustrates a broader distinction in agentic commerce: consumer authorization does not necessarily mean merchant authorization.
Agentic Shopping Is Still in Its Early Stage
For now, most AI shopping assistants are primarily being used for product research rather than autonomous purchasing.
Axios cited Coveo’s 2026 Commerce Relevance Report, which found that only 16% of shoppers were comfortable allowing an AI assistant to find and purchase products on their behalf. McKinsey estimates that U.S. retail revenue from AI-mediated search and purchase automation could eventually reach up to $1 trillion by 2030.
Retailers Are Taking Different Approaches
Amazon’s decision contrasts with Walmart’s approach.
Walmart has said it is working with Google and OpenAI to make Walmart products discoverable through their agentic shopping interfaces, even as Walmart develops its own shopping agent, Sparky.
Target, meanwhile, reported a 3.5x increase in traffic from external AI platforms over the previous year, showing that retailers may also see value in allowing AI platforms to influence product discovery.
The Next Phase of Agentic Commerce
The Amazon–Meta dispute shows that the future of agentic shopping will involve more than simply building smarter AI agents.
Retailers, AI companies and payment providers will need to establish mechanisms for agent identity, user consent, permissions, transaction limits, security and accountability.
The central question is shifting from “Can an AI agent shop for me?” to “Which agents are allowed to shop, where, under what permissions and who is responsible when something goes wrong?”
As agentic commerce develops, those rules could become as important as the AI capabilities themselves.
