Amazon blocks Meta's Muse agent, opening a fight over who controls the checkout when AI does the shopping

Amazon has blocked Muse, Meta's AI agent, when it tried to complete a purchase on Amazon's store. Users now see a notice that continued access by an unauthorized AI agent violates Amazon's terms of use. GeekWire reported the move and Yahoo Finance picked it up. Amazon says it first asked Meta to keep its site out of Muse's reach. Its argument: third-party apps that buy on customers' behalf should operate openly and respect each merchant's choice to take part or not. It compares the case to delivery apps and online travel agencies, which work under agreements with the merchants and airlines they transact with.

Meta's position, as reported by GeekWire, is that Muse never sees users' passwords or payment methods. Credentials users share go to secure storage. The agent runs on a dedicated virtual machine in Meta's cloud and generates single-use card numbers through Link by Stripe. A monitoring system called Sentinel controls what actions it can take online. The awkward part is that the two companies are partners. Amazon products have been buyable inside Facebook and Instagram since 2023. In April, Meta agreed to run AI workloads on Amazon's cloud infrastructure, according to Yahoo Finance.

The security argument is real, but the money explains more. Amazon earned more than $68 billion in advertising revenue last year, and that business depends on a human browsing the site and seeing sponsored listings. An agent that goes straight to the best product skips the ads, the recommendations and the pages designed to grow the cart. Amazon has done this before. According to GeekWire, it took legal action against Perplexity over its Comet browser and has also targeted shopping agents built by Google and OpenAI. Other sectors are closing doors too. Several airlines block outside agents during flight search or booking. JPMorgan Chase, Bank of America and Citigroup restrict automated bot access to customer accounts, officially for security and fraud prevention. In banking, an assistant that moves idle cash into higher-yield products would make deposits more expensive for the bank.

The infrastructure is shifting as well. Cloudflare Radar found that bots made up 60.6% of HTML requests in the seven days before August 10, 2026, versus 39.4% from humans, as reported by Search Engine Journal. Since September 15, Cloudflare's default settings block mixed-purpose crawlers, which combine search, training and agent tasks, on any page that shows ads, unless the site owner opts in, according to TechCrunch. The trouble is that the web cannot yet tell an agent acting with a user's permission from a malicious bot. The result is CAPTCHAs, dropped sessions and failed payments, and sometimes even human users get blocked.

Our reading: this is the 'plumbing war' we have described before. The fight is moving from who has the smartest model to who controls distribution and the point of sale. In the short term, expect walled gardens, lawsuits and private deals. Platforms will let agents in only when they get paid, much as travel agencies work with airlines. That is not entirely bad: Amazon's demand for agents to operate 'openly' is a reasonable governance principle. But if every merchant blocks by default, the user loses most, because the user is the one the agent is meant to serve. The likely way out is technical and regulatory: verifiable identity for agents, user-signed delegation, and open protocols that let a site know who is acting and on whose behalf. In the long run, agents that compare prices and move money for us push markets toward real competition and away from ad-funded attention capture. Incumbents will fight that shift, but it is the direction that cuts costs for everyone.