Amazon's Muse Block Is a Fight Over Who Owns the Checkout

Amazon runs the storefront, the warehouses and the delivery vans, and it has quietly become the landlord of online shopping. Picture a mall owner who controls the parking garage, the escalators and every register. He decides who may sell inside and collects a fee at the checkout.

Meta's new Muse agent tried to walk in. Users saw a pop-up on Sunday night saying it could no longer complete purchases on Amazon. The message was blunt: continued access by an unauthorized AI agent violates the Conditions of Use customers agreed to.

Amazon's objection is narrower than it sounds. Meta never told Amazon Muse would shop the store, the agent never identifies itself, and Amazon says Muse captures and stores customer credentials. Amazon also says it can reach account pages and order history.

The company reaches for a familiar comparison. A food-delivery app works with the restaurant rather than routing around it, and Amazon wants agents held to the same bargain. It says it is talking with Meta directly, and declined to say whether it will sue.

Meta tells a different story. Muse launched on September 8 as a personal agent that handles email, payments and shopping inside a secure virtual machine. Meta says Muse has no visibility into people's passwords or payment methods, and did not comment on the block.

Meta's launch materials describe the mechanism at the centre of the fight. Where a service offers a public API, Muse connects using credentials the user provides. Where it does not, the agent uses the service through a browser the way you would. A Verge reviewer linked her Amazon account, and Muse bought her tank tops.

Amazon runs its own agent, Buy for Me, which fetches items from outside brands' sites and lets those brands opt out. The objection, then, is not to agents but to unannounced ones.

The legal history explains the new language. Amazon sued Perplexity over its Comet browser and won a preliminary injunction in March 2026. The Ninth Circuit vacated that ruling in August, holding that the users do the accessing, not the agent.

Amazon has also narrowed its doors quietly, stripping item names and product images from order-confirmation emails and adding dozens of AI crawlers to robots.txt.

Every platform now wants to be the counter where agents check out. Visa, Mastercard, Stripe, Google and Meta are pushing competing protocols that cost merchants hundreds of thousands of dollars to integrate. Whoever owns that counter charges the fee, sees the data and decides what gets suggested.

For Amazon the prize is not the mug you buy but the auction for the slot that sold it. Advertising revenue hit $19.8 billion last quarter, up 26%, roughly $76 billion a year at margins far fatter than shipping boxes. Let an outside agent reach the register and Amazon becomes the warehouse while Meta keeps the sale.

Total revenue rose 20% to $200.6 billion, with operating income of $27.5 billion. AWS grew 37% to $42.2 billion and supplied $16.6 billion of that profit, while North America booked $116.2 billion at a 7.9% margin.

The capital spending is why that mix matters. Amazon spent $169 billion on property and equipment over the past year, and free cash flow for the twelve months to June was an outflow of $7.6 billion. Cash near $123 billion looks comfortable until you notice long-term debt, now about $129 billion, has roughly doubled this year.

So what has to happen for these shares to re-rate? Advertising has to keep compounding in the mid-twenties while AWS stays above 30%, and the spending has to become cash instead of capacity. The market pays for the first two and takes the third on faith, which is reasonable rather than proven.

The weak point is legal. The Ninth Circuit held that users do the accessing, which leaves conditions of use as the only fence around the store. No judge has tested that theory against an AI agent, and the answer decides whether the tollgate is an asset or a sign on the door.

There is a version of this where the bears are right. Amazon is spending like a utility to defend a tollgate no court has agreed it owns. The Ninth Circuit held that users do the accessing, so conditions of use are the only fence standing. A short seller would call the mismatch obvious: advertising carries the fat margin, while the retail segment the block protects earns a sliver. High-teens advertising growth in the next print is the number that proves them right.

The next checkpoint is Amazon's third-quarter print, where advertising growth against that 26% mark is the line that matters. The other clock is the courtroom. If the conditions-of-use theory holds, the tollgate becomes a moat. If it fails, the store is open again.

Disclosure: The Signal holds no position in AMZN. Positions may change. This is not financial advice.