Amazon blocks Meta's Muse agent from its marketplace

The dispute turns on how an AI agent presents itself when it acts for a shopper. Amazon says its terms require agents to identify themselves by embedding a text snippet in HTTP requests, and that Muse does not do so — and it adds that the agent can reach a user's Amazon account pages and purchase history, which it calls an undisclosed third party moving through customer accounts. Meta has said it is actively working to expand Muse's data protection guardrails, so it may seek to address those concerns to lift the ban.

Muse arrived with security machinery attached. Each Muse instance runs inside a virtual machine called the Muse Secure VM that isolates it from the rest of Meta's infrastructure, and a second agent, Sentinel, reviews sensitive actions such as purchases and keeps Muse from directly touching credit card numbers and account credentials. Even so, Meta can in theory reach data inside consumers' Muse instances today; the enhanced Muse Confidential VM, meant to make that impossible, is being tested with a limited number of users and is set to roll out later this year.

This is not Amazon's first clash of the kind. It sued Perplexity AI Inc. last year over the Comet browser and its shopping agent; an appeals court dismissed the case in August but did not bar Amazon from blocking agents that breach its terms of service. How the standoff ends likely hinges on whether Meta's coming confidentiality work and Amazon's identification rule can be reconciled — a question that matters to Meta's agent roadmap and to any retailer whose terms agents must follow.