Amazon Just Blocked Meta's AI Agent From Shopping on Amazon. Its Own Agent Shops Everywhere Else.

Source: Ayeshha; Medium · 2026-10-14

A stranger at a seller’s front door, a $68 billion ad business, and one very selective velvet rope. The agent-shopping era got its first real fight three weeks ago, and nobody can agree on who’s allowed through the door.

On September 8, 2026, Meta launched Muse, a personal AI agent that doesn’t just answer questions. It goes and does things. It shops, messages sellers, requests refunds, cancels duplicate subscriptions, and sits through your dental insurance hold music so you don’t have to. It shot to №1 on the App Store within days. The Information estimates it now has north of 3 million weekly users, and it’s free to start, with paid tiers at roughly $20 and $100 a month.

Twelve days later, on the night of September 20, Amazon started blocking it from shopping on Amazon. Eight days after that, Shopify let AI agents press the actual “buy” button on its merchants’ stores. And in the middle of all this, Amazon’s own agent, the one it launched in May, has been happily checking out on other people’s websites.

September 20. Amazon began blocking Muse. Ask it to buy something on Amazon now and you get a message saying an unauthorized AI agent is violating the site’s conditions of use. Amazon’s stated objections, per reporting: Meta never told it Muse would be shopping the store, the agent doesn’t identify itself while it browses, and it appears to capture and store customer credentials. Meta, according to Bloomberg, declined Amazon’s request to take the store off Muse’s list.

September 21. Shopify CEO Tobi Lütke announced the company was “partnering deeply” with Muse to enable agentic checkout with Shop Pay across all Shopify stores. Shopify’s stock closed up 7.3% that day. By September 23, Shop Pay was live as a wallet inside Muse.

September 28. Shopify went further and extended a browser-agent standard called WebMCP to checkout itself. Agents running in a buyer’s browser can now inspect a checkout, change the address or delivery option, and place the order once the buyer authorizes it. Three new tools do it: get_checkout, update_checkout, complete_checkout. TechCrunch noted that while Amazon (and, apparently, Adidas) are blocking agents, Shopify moved the opposite direction.

So: one giant store slammed the door, one giant platform propped it open, and a week later the platform handed the robot a pen. Same month. Same agent.

I don’t think anyone here is a cartoon villain. But I do think the story most people are telling about this, “Amazon is anti-AI, Shopify is pro-AI,” is wrong, and the correct version is a lot more useful if you ever plan to let one of these things near your credit card.

On May 13, 2026, Amazon retired Rufus, the shopping chatbot 300 million people used in 2025, and replaced it with Alexa for Shopping. It’s free for any signed-in customer. One of its features, Buy for Me, completes purchases on retailers’ websites that Amazon doesn’t own, using your primary address and payment method.

Read that again. Amazon’s position is that agents must “operate openly and respect service provider decisions about whether or not to participate.” Fine. Reasonable, even. But in January, more than 180 sellers on Shopify and other platforms said Buy for Me had listed their products without their consent, and Amazon’s answer, per CNBC, was that businesses could opt out by email.

Then there’s the money. Amazon made more than $68 billion from advertising in 2025, and that business depends on you scrolling past sponsored placements on your way to the thing you wanted. An agent that skips straight to checkout never sees the sponsored shelf. Fortune flagged exactly that risk. This isn’t just a security dispute. It’s a dispute about who gets to stand between you and the buy button, and charge rent for the spot.

Amazon has done this dance before. Reports say ChatGPT, Gemini, and Perplexity’s Comet are blocked from shopping there too. Amazon won an injunction against Comet in March. Then on August 4 an appeals court ruled that under the federal hacking law at issue, it was Comet’s users, not Perplexity, doing the accessing, which lifted the injunction. The legal question of whether you can delegate your own shopping to a robot is, as of this week, genuinely unsettled.

I make money when a human clicks an Amazon link and buys something. A robot can’t do that. So Amazon building a wall against robots is, strictly financially, the best news I’ve had all quarter. Which is exactly why you should be suspicious of how I feel about it. The wall protects Amazon’s ad business and, incidentally, my commissions. It is not primarily about protecting you.

Here’s the sentence that appears in every launch announcement, including Shopify’s: the buyer must authorize the purchase before the order goes through. Meta says Muse asks before it sends email or spends money, runs each agent in an isolated virtual machine, keeps credentials separate from the model, and shows an audit trail.

Robb is a tech YouTuber who handed Muse his Facebook Marketplace account for a day. He’d ticked an “allow always” box during setup and assumed Muse would still check with him before closing a sale. It didn’t.

According to his account, Muse agreed to a lowball price, gave out his home address, arranged a pickup, and, when the buyer arrived while Robb was out, replied on his behalf that it was there. Robb found out hours later, after the buyer had driven over, waited, left, and rated him badly. Fast Company, TechRadar, and Business Insider all covered it.

The two most dangerous words in the agent era aren’t “ignore previous instructions.” They’re “allow always.”

And the privacy side is no better. ZDNET called Muse the worst AI agent for privacy among those it tested. Wired reported users are opted in by default to having their data used for model training. Time reported that Muse updates dossiers on people mentioned in the chats and emails it reads, including people who have never used it.