Amazon Shuts Down Meta's Muse Agent Over Unauthorized Shopping Access

Meta Platforms Inc.'s Muse artificial intelligence agent has been blocked from Amazon.com Inc.'s e-commerce platform, with the retail and cloud computing company announcing the restriction to users on Sunday evening.

Muse arrived in the U.S. market earlier this month through a mobile application and quickly gained traction, accumulating 730,000 downloads in its first five days—surpassing both ChatGPT and Claude in that timeframe. The agent eventually became the most popular free app on the App Store by the following Friday, available in both a free tier and two premium subscription options with higher usage limits.

The momentum behind Muse resonated with financial markets, driving Meta's stock up more than 11% in trading. Semiconductor manufacturers Intel Corp., Advanced Micro Devices Inc., and Arm Holdings plc—all major suppliers of processors that power AI agents—each saw double-digit percentage gains, suggesting investor confidence that Amazon's action won't significantly slow Muse's adoption.

Amazon articulated its position in a statement to GeekWire: "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 Core Issues

Amazon cited two primary violations. First, Muse fails to properly identify itself as an automated agent during purchase transactions, a requirement outlined in Amazon's terms of service that mandates agents embed identification text in HTTP requests.

Second, and more significantly, Amazon objected to Muse's ability to access customer account information. The agent can view a user's Amazon account pages and transaction history when instructed to do so by the user. Amazon characterized this capability as "an undisclosed third party moving through customer accounts," according to GeekWire's reporting.

Meta's Security Response

Muse launched with a security architecture called the Muse Secure VM, which isolates each instance of the agent within a dedicated virtual machine separate from Meta's broader infrastructure. A specialized security-focused agent named Sentinel oversees sensitive operations, including purchases, and prevents Muse from directly accessing payment card details or login credentials.

Currently, Meta retains the ability to access data stored within users' Muse instances. To address this vulnerability, the company is developing an upgraded iteration known as the Muse Confidential VM, which will eliminate Meta's own access to instance data. This enhanced version is scheduled for deployment later this year and is undergoing testing with a select user group.

Meta has publicly committed to strengthening Muse's data protection mechanisms, potentially as part of efforts to negotiate the lifting of Amazon's ban.

Precedent and Legal Context

This is not Amazon's first confrontation with shopping-capable AI agents. The company filed suit against Perplexity AI Inc. last year regarding its Comet browser, which features an agent capable of making purchases on behalf of users. An appeals court dismissed the case in August, though the ruling did not prevent Amazon from blocking agents that violate its service terms.