Muse blocked from shopping on Amazon.com

What happened

Amazon blocked Meta's Muse personal AI agent from shopping on Amazon.com starting the night of 20 September 2026, after asking Meta to voluntarily exclude the retailer and getting no agreement. Users who try to buy something through Muse on Amazon now see a popup: "Continued access by an unauthorized AI agent violates Amazon's Conditions of Use, to which our customers have agreed." GeekWire first reported the block.

Amazon blocked Muse because Meta never disclosed that the agent would shop on Amazon.com, Muse does not identify itself when it browses, and it appears to capture and store customer credentials. Amazon says any agent that buys on a customer's behalf must operate openly and follow the same rules it applies to food delivery apps and online travel agencies booking on a customer's behalf.

What changed

• Amazon says Meta never told it in advance that Muse would access the Amazon store, and that Muse browses without identifying itself as a third-party agent, meaning an undisclosed party is handling customer accounts and payment data.
• An Amazon spokesperson told GeekWire and Adweek: "We think it's fairly straightforward that 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." The spokesperson added that Amazon has "requested that Meta remove Amazon from the experience."
• Amazon compared the obligation to how delivery apps work with the restaurants they order from, and how online travel agencies work with the airlines they book for customers: agents should be visible to the service they are using, not indistinguishable from a logged-in human.
• Meta launched Muse in the US on 8 September 2026 as its first consumer-facing personal AI agent. It runs on Muse Spark 1.3, a model Meta released on 2 September 2026 with a 1 million token context window that Meta says uses roughly 20% fewer tool calls and 25% less token usage on agent tasks than the prior Spark version.
• Each Muse user gets an isolated Muse Secure VM that holds the agent and the user's connected-app credentials, plus a separate Sentinel agent that checks every action leaving that VM against policies the user has already approved before letting it through. Meta's VP of Superintelligence Labs, Tarek Sheasha, has described the design as running in "its own isolated cell."
• Muse purchases go through single-use virtual cards issued via Stripe's Link, which Meta says means Muse's own model never sees a real card number.
• Amazon has drawn this line before: it sued Perplexity over its Comet agent allegedly making covert purchases, and has pushed back on shopping agents built by Google and OpenAI. A federal appeals court overturned an August 2026 ruling that had temporarily barred Perplexity's agentic shopping tools on Amazon.

Why it matters for agent builders

Anyone shipping an agent that acts on a retailer's site now has a public, named example of what gets an agent blocked: browsing without identifying itself, storing credentials the platform did not agree to, and skipping disclosure to the service it is transacting with. Amazon's stated bar, that an agent identify itself, offer an opt-out, and give the platform something in return, reads like a checklist a builder can test against before launch rather than after a block notice ships. The Perplexity precedent shows this can escalate to litigation, not just a popup.

What to watch

Whether Meta complies and removes Amazon from Muse's shopping scope, or keeps routing around the block, will show whether a company Meta's size treats a retailer's Conditions of Use as binding on an agent or as a negotiating position. It is also worth watching whether other retailers adopt Amazon's four-part standard, transparent operation, an opt-out, mutual value, and a better customer experience, as a shared baseline for which shopping agents they will let through.

Key takeaways

• Amazon blocked Meta's Muse from shopping on Amazon.com on the night of 20 September 2026, citing undisclosed access, credential capture, and an agent that does not identify itself.
• Amazon asked Meta to remove Amazon from Muse's shopping experience rather than filing suit, unlike its earlier action against Perplexity's Comet agent.
• Muse runs on Muse Spark 1.3 inside a per-user Muse Secure VM, with a separate Sentinel agent gating every action that leaves the VM.
• Muse purchases use single-use virtual cards from Stripe's Link rather than exposing real card numbers to the agent or the merchant.
• A federal appeals court had already overturned a ruling that temporarily barred Perplexity's shopping agent from Amazon, so the legal boundary here is still moving.

FAQ

Why did Amazon block Meta's Muse agent?

Amazon says Meta never disclosed that Muse would shop on Amazon.com, that Muse does not identify itself as a third-party agent while browsing, and that it appears to capture and store customer credentials without Amazon's knowledge or consent.

What is Muse Secure VM and Sentinel?

Muse Secure VM is an isolated, per-user virtual machine that runs the Muse agent and holds a user's connected-app credentials. Sentinel is a separate agent that checks every action trying to leave that VM against policies the user already approved, and stops to ask if nothing matches.

Has Amazon blocked other AI shopping agents before?

Yes. Amazon sued Perplexity over its Comet agent allegedly making covert purchases on behalf of users, and has also pushed back on shopping agents built by Google and OpenAI. A federal appeals court overturned an August 2026 ruling that had temporarily barred Perplexity's tools on Amazon.

Source: GeekWire

Discussion: Hacker News

Builders comparing commerce and payments agents against this new bar can browse the payments and commerce category on aiagentslisting.com.