Amazon blocks Meta's Muse AI agent from shopping on its store

Why it matters

Agentic commerce depends on access to websites owned by potential competitors. Amazon's block shows that platform rules, contracts and control of the checkout may constrain consumer agents before their technical capabilities do.

Amazon blocked Meta's Muse AI agent from shopping on Amazon.com on Sunday night, September 20th, cutting off one of the assistant's core use cases 12 days after Meta launched it.

Amazon told GeekWire that Meta had declined a request to exclude Amazon from Muse. Customers who try to send the agent shopping on Amazon now see a warning that access by an unauthorized AI agent violates the retailer's Conditions of Use.

Amazon says Meta's Muse browses without identifying itself, can enter customers' account pages and order histories, and relies on stored credentials to act for users. Amazon argues that outside agents should identify themselves and give service providers the choice to participate.

Meta's account is narrower. In its September 8th launch announcement, Meta said Muse cannot see passwords or payment methods. Credentials are stored separately and inserted into the browser when needed, according to Meta, while users must approve sensitive actions such as purchases.

Muse was built to cross platform boundaries

Meta designed Muse to complete multi-step tasks across services rather than remain inside Meta's own apps. Muse has a browser, can fill out forms and keeps working after a user closes the app. When a service lacks a public integration, Meta says the agent can operate through its website as a person would.

That design makes platform conflict a built-in feature. An agent that can shop, book travel, manage email and negotiate bills becomes useful by moving across properties controlled by other businesses. Those businesses still set the rules for their websites, customer accounts and transaction systems.

Shopify co-founder and CEO Tobi Lutke publicly drew the contrast, saying Shopify does the opposite. The split leaves commerce platforms making different choices about whether outside agents require prior permission.

Meta says Muse runs inside a dedicated virtual machine with separate systems for credential storage, network permissions and browser activity. Its technical description of Muse's security architecture says login information is routed directly into protected storage outside the agent's runtime environment. The agent receives an accessibility snapshot of a page rather than the raw document structure, and it pauses when a user takes over the browser or enters credentials.

Those safeguards address whether Muse itself can read or leak a password. Amazon's objection reaches further: Amazon says Meta deployed an outside system inside customer accounts without Amazon's consent or a machine-readable identity that would let Amazon distinguish Muse from a person browsing normally.

Amazon changed its legal route after losing to Perplexity

The wording of Amazon's warning reflects a recent court loss. Amazon previously sued Perplexity over shopping activity conducted through its Comet browser and initially obtained a preliminary injunction.

On August 4th, a Ninth Circuit panel vacated that injunction. The court found that Amazon was unlikely to prevail under federal and California anti-hacking laws because the customer, rather than Perplexity, was accessing Amazon with an AI tool. The court denied Amazon's rehearing petition on September 10th.

The ruling preserved Amazon's ability to regulate access through private contracts and terms of service. Amazon's message to Muse users accordingly cites its Conditions of Use instead of accusing Meta of unauthorized computer access under anti-hacking law.

That distinction will shape similar fights. AI developers can argue that an agent acts as software under a customer's direction. Platforms can respond that users agreed to contractual limits covering automated access, account information and purchasing behavior. Amazon's Muse block is an immediate enforcement test of that contractual approach.

Amazon is defending the shopping interface

Security and transparency are central to Amazon's stated position, but the dispute also concerns control of product discovery. Amazon reported $68.6 billion in advertising services revenue for 2025 in its annual filing. Sponsored products depend on customers searching, comparing and browsing inside Amazon's interface.

An outside agent can compress that journey into a request and a purchase approval. That could reduce the pages a shopper visits, the sponsored listings a shopper sees and Amazon's ability to control how products are ranked and presented. Amazon has a direct commercial interest in keeping the agent layer inside products it operates.

Amazon already offers its own AI shopping tools. Alexa for Shopping operates inside Amazon's retail experience, while Buy for Me searches external brand websites. Amazon says Buy for Me identifies itself to those merchants and allows them to opt out, the same conditions Amazon says Meta failed to provide with Muse.

The confrontation is confined to the storefront. Meta and Amazon remain substantial business partners elsewhere. Amazon products have been available for purchase through Facebook and Instagram since 2023. In April, Meta also agreed to deploy tens of millions of AWS Graviton cores for AI workloads, including the multi-step coordination used by agents.

Amazon is willing to supply the infrastructure behind Meta's agents while refusing to let one of those agents enter its store without permission. That boundary shows where cooperation ends: Amazon can sell Meta the computing power, while Amazon keeps the checkout, customer relationship and advertising inventory under its own control.