By Remco Livain in AI — 24 Sep 2026

Amazon blocked Meta's AI shopper. It's not about security.

Amazon cut Meta's Muse agent off its marketplace, citing security. The real threat is to the browsing, ads and impulse buys the marketplace runs on, and what every online shop should do about agentic commerce now.

Over the weekend of September 20, anyone who asked Meta's new Muse agent to buy something on Amazon hit a wall. Not an error, a warning: continued access by an unauthorized AI agent violates Amazon's Conditions of Use. GeekWire broke the story, and Amazon's stated reasons are all about security. The agent doesn't identify itself, it didn't ask permission, and it appears to handle customer credentials.

Those concerns are real. I just don't think they're the reason. I recorded my take on it this week:

An agent doesn't browse

Think about what an AI agent actually does on a marketplace. You tell it what you need, it finds the cheapest version that fits, it buys it, and it leaves. No scrolling. No "customers also bought". No sponsored listing in position two. No second item in the basket because something caught your eye on the way to checkout.

That browsing is the business. Amazon's advertising arm alone brought in more than $68 billion last year, and every dollar of it depends on a human being looking at a page. An agent that shops efficiently is, from Amazon's side of the table, a customer who never sees an ad and never adds the impulse item. That's not a security problem. That's a business model problem.

"I constantly remind our employees to be afraid, to wake up every morning terrified. Not of our competition, but of our customers." Jeff Bezos, Invent and Wander

I highlighted that line years ago, and it reads differently this week. The customers are the ones asking Muse to do the shopping. Blocking the agent is blocking a customer who has decided to shop differently.

We've seen this movie before

When DoorDash and Uber Eats arrived, restaurants had the same fear. A middleman would sit between them and their guests, take a cut, and own the moment where people decide where to eat. Discovery would move somewhere they didn't control. Some restaurants fought it. Most ended up on the platforms anyway, because that's where the customers went.

The same shift is happening one layer up. More and more, a product gets found by being the answer an LLM gives: in ChatGPT, in Gemini, in Muse. Where the product is listed and where it gets bought starts to matter less than whether the assistant picks it. My guess is that within the next 24 to 36 months a meaningful share of people will simply say "buy this for me" to whatever assistant they already use, and never visit the shop at all. I wrote about the other side of that coin when Cloudflare started letting sites split AI training from AI answers: the question is no longer whether bots reach you, but which ones you let in and on what terms.

Amazon already has its own agent

What makes this more interesting is that Amazon isn't against agents. It's against agents it doesn't own. In May it turned Rufus into Alexa for Shopping, an agentic assistant that researches, recommends and buys for you inside Amazon's walls. Inside its own ecosystem, "just buy it for me" is fine.

Muse isn't the first outside agent to be turned away, either. Amazon has been tightening its robots.txt against AI crawlers for a while, including the bots ChatGPT uses to read live pages, and it took Perplexity's Comet agent to court. So the real question is what comes next. When Gemini, Claude, ChatGPT and Grok all ship agents that can buy, does Amazon block every one of them?

Partner, don't block

I don't think blocking is a winnable game. Agentic shopping isn't going away, and Amazon has exactly what these assistants need most: a trusted product database, real reviews, real prices, shipping weights, delivery dates, and logistics nobody else can match. That's the part of retail they have nailed. The stronger move is to be the fully integrated layer every assistant relies on, not the store that turns them away at the door.

It's also a pattern I see with clients all the time. The instinct under pressure is to defend the model you have. The better question is where you still own the customer relationship once the interface changes, and how to make yourself indispensable there.

What this means if you sell anything online

You don't have to be Amazon for this to matter. Every shop is about to be read by bots before it's seen by people. Can an assistant understand what you sell, at what price, and when you can deliver it? Is your product data structured, or buried in a PDF price list? Do you let the crawlers that matter pick up your products and your prices, or does a default setting keep them out?

Data clarity for bots is going to decide who gets recommended over the next couple of years. The shops that sort it out now will profit while the big players are still arguing about terms. This is where experience matters more than tooling. Any AI will happily write you a "get ready for agentic commerce" plan. Knowing which three fields in your feed actually decide whether you get picked is what twenty years of doing the work buys you.

Amazon can close the door on Muse. It can't close the door on the way people are starting to shop.

Sources & further reading

External
GeekWire: Amazon blocks Meta's Muse AI assistant in new standoff over agentic shopping
TechCrunch: Meta's AI agent has been blocked from using Amazon.com
Forbes: Amazon's $68 billion reason to block Meta's Muse
About Amazon: Meet Alexa for Shopping
Modern Retail: Amazon quietly blocks more of OpenAI's ChatGPT web crawlers
Jeff Bezos, Invent and Wander (Harvard Business Review Press, 2020)