Amazon Blocked Meta's Shopping AI and Gave Sellers a Free One. Connect Yours Before Q4.

Amazon did two things four days apart in September that look contradictory and are not.

On September 21, it locked Meta’s new AI shopping agent, Muse, out of its store, telling users the app was violating Amazon’s Conditions of Use. On September 23, it announced that sellers can now run their entire Amazon business through Anthropic’s Claude, free, with no coding required.

The recommendation: stop thinking of this as Amazon versus AI agents. Amazon is fine with agents. It is not fine with agents it did not authorize acting on customers, and it is actively building the authorized version for sellers. Connect the Claude plugin this week. It is free through the end of the year, it took beta testers about a minute to set up, and it is the closest thing to a free ops hire you will get before peak.

What actually happened, in order

Meta’s Muse launched September 8 and hit number one on Apple’s U.S. App Store within a week, ahead of ChatGPT. It let users hand over shopping tasks, including purchases on other retailers’ sites, using stored payment credentials.

Sunday night, September 21, users trying to shop on Amazon through Muse got a popup: “Continued access by an unauthorized AI agent violates Amazon’s Conditions of Use, to which our customers have agreed.” Amazon’s public line, per GeekWire’s reporting: “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.” Meta’s response was that Muse never sees a user’s password or payment method directly, that credentials sit in secure storage the agent can use without seeing.

Two days later, September 23, Amazon announced that Seller Assistant, its existing AI tool for sellers, now runs as a plugin inside Amazon Quick and inside Anthropic’s Claude. A seller can check inventory, adjust prices, update listings, and diagnose ad campaigns from inside Claude without logging into Seller Central at all. Setup takes about 60 seconds. Mary Beth Westmoreland, Amazon’s VP of Worldwide Selling Partner Experience, put the intent plainly: “Our vision was that they would never have to log into Seller Central. We would just bring it to them where they work.”

This is not Amazon’s first fight over agentic shopping. It sued Perplexity over its Comet browser, won a preliminary injunction in March, then lost it on August 4 when the Ninth Circuit ruled that under federal anti-hacking law, it is the user accessing the website, not the AI company. That ruling is exactly why Amazon is leaning on its Conditions of Use and account-level enforcement against Muse instead of a lawsuit this time, and why it will keep doing this to the next agent that shows up uninvited.

Why blocking one and opening the other is the same decision

Amazon runs two businesses that both depend on controlling what happens between “I want this” and “I bought this.” One is fulfillment and logistics. The other is advertising, which grew 26% year over year in Q2 2026 according to Amazon’s own earnings report, on a $200.6 billion sales quarter.

An unauthorized agent that logs in as the customer, browses the catalog, and checks out is invisible to both systems. It does not see Sponsored Products. It does not get attributed the way a normal session does. It routes around the exact mechanics that make Amazon’s ad business and delivery-promise engine work. That is a direct threat to the two things Amazon monetizes best, which is why the response was immediate and public rather than a quiet technical fix.

An authorized agent that a seller invites in to manage their own account is the opposite. It keeps that seller’s operations, listings, and ad spend inside Amazon’s ecosystem instead of migrating to a third-party tool. It deepens Amazon’s relationship with Anthropic, whose Claude models already sit under Seller Assistant. And it costs Amazon nothing to give away, because the seller was already going to spend hours a week in Seller Central regardless of which AI helps them do it.

Same logic, opposite door. Agents that reduce Amazon’s leverage get blocked. Agents that extend it get shipped as a feature.

The math on what Amazon just gave you

Amazon Quick Plus, the tier the Claude and Quick plugin runs on, lists at $20 to $25 per user per month depending on billing term. Amazon is giving every primary seller account a free 12-month Quick Plus subscription through December 31, 2026, plus complimentary seats for two team members.

Run the number on a lean team of three:

• 3 seats x $20/month (annual rate) x 12 months = $720 in software you are not paying for
• 3 seats x $25/month (monthly rate) x 12 months = $900 if you would have paid month to month

That is not a rounding error going into Q4, and it is not the real value. The real value is what the plugin actually does once it is connected: it holds memory of your pricing patterns, inventory cycles, and growth goals across sessions, it runs workflows continuously in the background with guardrails you set in plain language, and it produces a full audit trail of every action it takes before you approve it. Amazon reports sellers already accept its recommendations more than 90% of the time. This is not a chatbot bolted onto Seller Central. It is closer to a junior operations hire who never sleeps, never asks for a raise, and reports every move it makes.

What to do this week

Connect the plugin, do not just read about it. Beta access is U.S.-only for now, through Amazon Quick and through Claude. If you use Claude already for anything else in the business, this is a five-minute add, not a new tool to learn.

Start it on the boring stuff, not the risky stuff. Account health monitoring, order summaries, and pricing analysis are the prebuilt templates Amazon ships. Let it watch and report before you let it act. Every action still requires your approval unless you explicitly grant standing permission, so there is no reason to skip the trial run.

Audit anything in your stack that already claims to “shop for you” on marketplaces. If an agency, a script, or a tool you use logs into Seller Central using your credentials to act like a customer or automate purchasing behavior rather than using Amazon’s sanctioned APIs, treat the Muse block as the preview of what happens to it. Amazon has now shown twice this year, once in court and once by popup, that it will cut off access it did not explicitly authorize. Unauthorized workarounds are a liability on your account health, not a growth hack.

If you also run a DTC storefront, do not assume the Ninth Circuit ruling protects you the way it does not protect Amazon. The same logic that let Muse legally access Amazon’s site as “the user” applies to any AI shopping agent hitting your Shopify checkout. If you have not decided whether you want agents transacting on your own site, and on what terms, that decision is coming for you next, not just for the marketplaces.

The read for Q4

Amazon is not choosing between AI agents and control. It is building the version of agentic commerce it controls, on both sides of the transaction, seller tooling and shopper tooling, and closing the door on every version it does not. That is a rational, well-funded strategy, and it means more of these announcements are coming, not fewer.

The sellers who get value out of this are not the ones debating whether AI agents are good or bad for ecommerce. They are the ones who connected the free tool in September and had three months of clean data and workflows running before anyone else started.

If you want an operator team running your Amazon, Walmart and Shopify accounts who catches moves like this before they hit your Q4 numbers, a strategy assessment is the fastest way to get started.