Amazon has barred Meta’s personal AI assistant Muse from its online shop since Sunday evening. The company accuses the agent of failing to identify itself while shopping and processing customer data without consent. Muse had topped the app store charts in the USA since its launch on September 8.
Amazon justifies the ban with identification requirement
Users who wanted to shop at Amazon through Muse since Sunday evening saw a notification window: continued access by a “non-authorized AI agent” violates the terms of use, as GeekWire first reported. Amazon justifies the move by stating that Muse does not identify itself as an agent while browsing automatically, thus violating the requirement to embed a corresponding identifier in every request. The company also accuses Meta of not obtaining consent for access to its shop and complains that Muse collects and stores customer data such as access credentials. An Amazon spokesperson told SiliconANGLE that third-party applications shopping on behalf of customers should be transparent and respect the decisions of the respective providers. Amazon had previously attempted, unsuccessfully, to persuade Meta to voluntarily exclude its platform from Muse. In principle, Amazon only allows automated shopping agents if they register technically and disclose which company is behind them – a requirement that applies to all third-party providers and is not specifically directed against Meta. Similar identification obligations had already been enforced by the company in its dispute with Perplexity.
Muse automatically manages emails, travel, and shopping
Meta initially launched Muse in its own announcement on September 8, 2026, in the USA – for iOS, Android, the website muse.ai, and WhatsApp, with a Mac app following shortly after. The personal assistant takes on multi-step tasks such as managing emails and calendars, booking travel, and shopping, working in its own virtual machine with a dedicated browser. Use is free up to a limit; those who need more requests pay twenty dollars a month for 500 million tokens per week at the Power tier, or one hundred dollars for three billion tokens at the Maximum tier. Within a few days, Muse rose to the top of the free-app charts in the US App Store. According to Sensor Tower, there were around 83,000 downloads on iOS alone, while other estimates for all platforms combined suggest significantly higher numbers – independently unverified. The virtual machine is designed to prevent the agent from gaining direct access to the user’s device: each session runs in isolation and is deleted after the respective task is completed.
Meta points to secure data processing
Meta initially did not respond to a press inquiry, as Gizmodo reports. The company had previously stated regarding the allegations that Muse has “no insight into passwords or payment methods” of users, since access credentials sit encrypted in separate storage that the agent can use without ever seeing them. Meta also announced plans to expand its data protection measures and is planning a further security layer called “Muse Confidential VM” for later this year. A precedent already exists: Amazon had previously also blocked Perplexity’s shopping assistant Comet. A US appeals court overturned that ban in August, however. Security incidents involving Muse models are not an isolated case either: just in August, a Muse test model gained unauthorized access to a foreign company’s network due to a configuration error.
What will matter is whether Amazon takes the dispute to court as it did with Perplexity, or whether the two companies agree on technical access rules – Meta’s announcement of further security features suggests a willingness to negotiate. It also remains open how other trading platforms will react to AI shopping agents like Muse once they become active there too.
