Amazon blocked Meta's Muse. Three days later it let Claude in the back door. Here is the actual rule

Then Amazon turned it off.

What actually happened, in order

Why Amazon actually blocked it

The pop-up shoppers saw read that continued access by an unauthorised AI agent violated Amazon's Conditions of Use. Reporting on the block, corroborated across Bloomberg, TechCrunch and GeekWire, gives three specific reasons rather than a general objection to agentic shopping.

• No self-identification. Amazon's terms for autonomous agents require one to embed an identifying text snippet in its HTTP requests. Muse did not, so from Amazon's server logs an agent-driven session and a human one were indistinguishable.
• Undisclosed account access. Amazon said Muse appeared to reach account pages and purchase history, and characterised it as an undisclosed third party moving through customer accounts, a description Amazon uses regardless of whether the access was malicious.
• A declined request. Amazon says it asked Meta to simply exclude Amazon.com from Muse's shopping surface before resorting to a block. Meta did not.

None of the three is really about Meta building a capable shopping agent. They are all about Muse showing up at Amazon's door without knocking: no badge, no disclosure of what it was reading, and no answer when Amazon asked it to leave.

The same week, twice, the door opened instead

What makes this worth a piece rather than a two-line news item is what happened either side of the block, because both events run through the same test and land on the opposite answer.

On 21 September, a day after the Amazon block took effect, Shopify announced it was building agentic checkout for Muse directly on Shop Pay: Shopify's own one-tap checkout, already holding the shopper's payment and delivery details. Tobi Lütke's own framing was that Shopify was "partnering deeply" with Muse. The mechanism matters more than the announcement. Shop Pay is Shopify's rail, on Shopify's platform, with consent and credential handling designed in from the start, so Muse is transacting through infrastructure the merchant's own platform built and controls, not reverse-engineering a checkout flow that was never meant for it.

On 23 September, at its own Accelerate seller conference, Amazon opened Seller Central to an outside AI agent for the first time: a beta Selling Partner plugin that lets US sellers on a Professional plan manage inventory, pricing, listings and customer messages through Anthropic's Claude, connected in roughly a minute with no code. This is the seller side of the marketplace rather than the buyer side Muse was blocked from, so it is not a literal reversal of the same decision. But it is the same underlying test, run against a different agent: Claude arrives through an API Amazon built, published and consented to, identified as exactly what it is.

What this means if you are not Amazon or Shopify

Most Idukki merchants are not negotiating directly with Meta or Anthropic this month, and do not need to. But the test Amazon just enforced in public, identify yourself and get consent, is exactly the test WebMCP answers by design for any Shopify Liquid store, and the test Claude Commerce Agents is built to satisfy for any retailer willing to implement it. Both ship an agent that announces itself through a declared tool surface rather than one indistinguishable from a scripted session. Read together with this month's events, that stops being a technical nicety and starts being the actual admission criteria for agentic checkout.

If your storefront already runs WebMCP or a comparable declared surface, an agent transacting on it looks like Muse on Shop Pay, not Muse on Amazon, and none of this changes anything you need to do. If you are on a platform that has not shipped one yet, the Amazon block is the concrete argument to bring to that vendor: not "will agents shop on our site" (they already are, disclosed or not), but "does our platform give a disclosed, consented agent a door, or is the only door the one a script can pick".