Meta’s Muse can assist with shopping, but it cannot currently complete purchases on Amazon. Users who tried to buy goods through the AI assistant on Sunday night began receiving an error saying that continued access by an unauthorized AI agent violated Amazon’s conditions of use.
The block turns a broad debate about agentic commerce into a specific platform boundary. A consumer may choose an AI assistant, and that assistant may be technically capable of navigating a purchase, but the retailer still controls whether the software is allowed through the door.
A Shopping Agent Meets a Platform Rule
The error was spotted by GeekWire and reported by TechCrunch. It identifies Muse as unauthorized rather than describing a temporary outage or failed checkout. For affected users, the practical result is straightforward: a purchase initiated through Muse must be taken to another retailer or completed without the agent.
No negotiation, integration plan or timetable for restoring access was described in the report. The verified development is therefore limited to the block and Amazon’s stated conditions-of-use rationale. It does not establish that Amazon has rejected all AI-mediated shopping permanently.
Even so, the incident shows that an agent’s usefulness depends on cooperation from services it does not control. Meta can design a capable interface and attract users, but it cannot unilaterally grant that interface permission to transact on another company’s marketplace.
That dependency complicates the idea that a general assistant can shop anywhere on a user’s behalf. Each merchant can decide which automated visitors it recognizes, how they authenticate and what actions they may take.
Rivalry Is Only One Possible Explanation
The companies have reasons to view each other competitively. Amazon operates its own group of foundation models and a widely used inference platform. TechCrunch notes that, absent a legal duty to provide access, Amazon has little obvious incentive to open its store to a rival company’s agent.
But the report also identifies a more operational concern. If Muse places an incorrect order, Amazon may have to manage the consequences for both the customer and the seller. Returns, complaints and disputed purchases would arrive at the marketplace even when the mistaken decision originated in Meta’s system.
Muse is described as having a comparatively low hallucination rate, but not a zero rate. The source does not provide a measured error rate for shopping tasks, nor does it say that a faulty Muse order caused the block. Liability and support costs are presented as plausible reasons for caution, not confirmed motives from Amazon.
That distinction prevents the episode from being reduced to a feud between large technology companies. Competitive positioning may matter, while ordinary transaction risk may be enough to justify a restrictive policy.
Agentic Commerce Needs Permission and Accountability
The block exposes two separate requirements for automated shopping. An agent must be able to find and select products accurately, and the destination marketplace must be willing to accept the agent’s actions. Progress on the first problem does not solve the second.
The report suggests Amazon could still see an opportunity in agent-driven commerce while choosing to wait through additional product releases. Waiting would allow the technology to mature before the marketplace takes on the customer-service burden created by automated mistakes. That is a possibility raised by TechCrunch, not a publicly announced Amazon roadmap.
For Muse users, the present limitation is narrower and more immediate: Amazon shopping is unavailable through the assistant. For the wider market, the episode demonstrates that autonomous purchasing will be governed as much by access rules and responsibility for errors as by model capability. Agents may promise a seamless route from request to checkout, but retailers retain the power to decide whether that route exists.
