Amazon is blocking Meta’s new AI agent Muse from accessing its US online store. Users who tried to have Muse find or order products saw a notice about an unauthorized AI agent. The dispute reveals a limit to automated shopping: an assistant may be able to plan and click, but it cannot complete a purchase if the store will not accept it as a customer channel.
Key takeaways
- Amazon has blocked Muse from shopping on Amazon.com, according to multiple reports, citing a lack of authorization.
- Amazon points to transparency, account security, and the shopping experience; Meta initially did not respond to Axios’ request for comment.
- The incident concerns one agent’s access to Amazon’s store, not every Muse feature or every retailer.
- For shoppers, the key questions are who handles bad orders and who controls comparisons, data, and checkout.
What was blocked
Muse is Meta’s personal AI agent, offered in the United States since September for everyday tasks. According to Meta, it can operate a browser, fill out forms, and help with shopping. Our earlier overview of Muse covered that new capability and its initial US-only availability. Now a specific part of the promise has run into a major retailer: when users tried to shop on Amazon through Muse, TechCrunch and others reported, they saw a message saying that continued access by an unauthorized AI agent violated Amazon’s Conditions of Use.
Amazon told Axios that it had not authorized Muse to access customer accounts, scrape store data, or process transactions. The company says third-party applications that shop for customers should identify themselves and respect a service provider’s decision about participation. Meta did not initially answer Axios’ request for comment. That does not establish exactly how the block works: the reports document an obstacle in the shopping flow, but they do not show which detection method Amazon uses or whether other Meta features are affected.
For readers outside the United States, the restriction is mainly a preview of an emerging market conflict. Meta initially launched Muse only in the US. The news therefore does not mean a previously available Muse shopping feature has disappeared in Europe. It still matters because retailers and agent providers everywhere will need to decide when software acting for a customer may operate a customer account.
Why Amazon objects
An AI agent changes the roles in a purchase. Normally Amazon sees how a customer searches, compares products, reads reviews, and completes an order. An outside assistant can combine those steps and shape the choice before the customer reaches the store. That sounds convenient for buyers: one instruction instead of many browser tabs. For the retailer, it raises questions about the identity of the software, the handling of login credentials, and who must help when an agent chooses the wrong size, an unsuitable product, or an unintended order.
Amazon cites customer and data security, as well as the quality of the shopping experience, in its comments to Axios. Those are plausible concerns, but they do not tell the entire business story. Whoever mediates product search also influences which offers are visible and who learns about purchasing intent. Amazon has its own AI shopping functions, giving it an incentive to preserve its direct relationship with customers. Both considerations can be true at once: a real operational and security question, and a conflict over the future interface to commerce.
The position predates Muse. In a letter to Perplexity published in 2025, Amazon said that the company’s Comet shopping agent needed to identify itself and respect Amazon’s decision on automated access. That document concerns a different provider and does not establish any particular allegation against Meta. It does show that Amazon regards transparency and consent as general conditions for outside shopping agents.
What buyers and providers can learn
Anyone using a shopping agent should distinguish research from placing an order. A system that summarizes offers can save time even if it does not pay on the user’s behalf. Checkout adds more: shipping address, payment details, returns, seller terms, and whether a click truly reflects the user’s intent. A dependable assistant should show the price, seller, delivery date, and total cost clearly before purchase and ask for unmistakable approval. That is a practical product requirement, not a claim that Muse mishandled those steps in this case.
For agent providers, working browser controls are not enough either. They need agreements or other accepted interfaces with stores, clear identification rules, and a credible way to handle mistakes. Retailers cannot simply dismiss demand for convenient product discovery. They must decide which agents may enter under which conditions while protecting customer accounts and checkout processes. An open, documented interface could ease the conflict, but no such solution for Muse and Amazon has been announced.
The next barrier is organizational
The block says little about the technical capability of Meta’s model. It shows that the next obstacle for AI agents often sits outside the model: permissions, commercial relationships, and responsibility for a transaction. Users will get lasting convenience only if assistants and retailers agree on a route to purchase. Until then, Muse may help plan a purchase, but Amazon decides whether the agent may take the final step in its store.