No Sign-In Sheet for Agents: Amazon Blocks Meta's Muse From Shopping Its Store

Late on Sunday night, Muse users who tried to send Meta’s new personal AI agent to Amazon.com to buy something hit a wall. A popup warned that “continued access by an unauthorized AI agent violates Amazon’s Conditions of Use, to which our customers have agreed.” By Monday morning, Amazon had confirmed it: Meta’s Muse — the personal AI agent launched on September 8 that can browse, select and check out on a user’s behalf — is no longer welcome to shop the world’s largest e-commerce platform.

It is the sharpest public clash yet in a fight that has been building for over a year: who controls the online shopping experience when the shopper is no longer a human, but an AI agent acting for one.

Amazon’s three complaints

According to GeekWire, which first reported the block, Amazon’s objections fall into three buckets.

First, no notice. Meta never told Amazon that Muse would be accessing its store. The retailer learned about it the way everyone else did — when users started dispatching the agent to buy things. Second, no identification. Amazon says Muse does not identify itself as an automated agent while browsing, meaning the store cannot distinguish an agent’s traffic from a customer’s. Third, and most seriously, credential capture. Amazon says Muse appears to capture and store customer login credentials, which introduces security risks it never agreed to.

“We think it’s fairly straightforward that third-party applications that offer to make purchases on behalf of customers from other businesses should operate openly and respect service provider decisions about whether or not to participate,” an Amazon spokesperson said in a statement.

Meta’s defense

Meta’s position is that Amazon is mischaracterizing how Muse works. The company says Muse has no visibility into passwords or payment information; user credentials are placed into secure cloud storage so the agent can execute purchases without ever seeing them. When Meta launched Muse earlier this month, it emphasized that the agent cannot read secure login details or card data, and that purchases require explicit user approval.

The two companies are, awkwardly, still business partners — including a multibillion-dollar AWS deal that hosts Meta’s agentic workloads. Amazon says it is in direct discussions with Meta about the assistant, but declined to say whether it plans to escalate to a lawsuit. Meta did not immediately respond to requests for comment Sunday night.

Why Amazon changed its legal strategy

The Conditions-of-Use popup is not Amazon’s first move against AI shopping agents — but it is a deliberately different kind of move. Last November, Amazon sued Perplexity to keep its platform out of the Comet browser’s shopping experience, and won a preliminary injunction in March. That win evaporated on August 4, when the Ninth Circuit dissolved the injunction, ruling that under federal anti-hacking statutes it is consumers — not the AI companies building tools for them — who are accessing the computers in question. A petition for rehearing was denied on September 10.

That loss matters. Amazon’s statutory hacking claims against agent companies are, for now, dead ends. So the company has pivoted from statute to contract: instead of arguing that agents “hack” Amazon, it argues that customers agreed to Conditions of Use that prohibit unauthorized automated access, and that violating those terms puts accounts at risk. It’s a quieter strategy, but one that doesn’t depend on persuading the Ninth Circuit.

Amazon has also been quietly hardening its storefront against agents in general. Confirmation emails have been notably sparse since July, omitting specific item names and product images — reportedly to limit how much external AI services can mine from them. And the company has spent the past year blocking external agent tooling, including crawlers and shopping helpers from OpenAI and Google, from reading its product pages.

Follow the advertising money

The commercial stakes are enormous. Amazon generated more than $68 billion in advertising revenue last year, and that revenue is tied directly to shoppers seeing search results pages and sponsored product listings. An agent that skips the browsing experience, compares prices invisibly, and buys whatever best matches the user’s stated goal doesn’t see — or care about — sponsored placements. Agentic shopping threatens to unbundle Amazon’s highest-margin revenue stream from its marketplace.

There is also a double standard at play that Amazon is acutely aware of. Earlier this year Amazon launched “Buy for Me,” its own feature that finds and purchases products on external websites. Amazon notes, correctly, that Buy for Me identifies itself to merchants and honors their opt-outs — exactly the openness it says Muse lacks. The message is implicit but clear: agents shopping on Amazon’s turf must follow rules that Amazon’s own agent follows when shopping on everyone else’s.

The road ahead

For now, the standoff is diplomatic rather than legal. But the shape of the dispute is a preview of the next decade of e-commerce. Personal agents from Meta, OpenAI and Google want to become the interface through which consumers buy everything — disintermediating the stores themselves. Retailers want to keep the customer relationship, the browsing data and the ad inventory. The Ninth Circuit has taken one weapon off the table; contracts, popup warnings and quiet infrastructure changes like stripped-down emails are the new front line.

The uncomfortable question neither company answered this weekend: if an agent acts with a user’s consent, using the user’s own account, whose permission actually matters — the user’s, or the platform’s? Amazon has answered with a popup. The industry still doesn’t have a better one.