Amazon Blocks Meta’s Muse Shopping Agent Days After U.S. Launch

Amazon has blocked Meta’s Muse personal AI agent from shopping on Amazon.com, turning a fast-rising consumer assistant into an early test of who controls agentic commerce. The dispute centers on whether a user-authorized AI agent can browse, handle credentials and buy from a merchant that has not agreed to participate. Meta introduced Muse on Sept. 8 as a personal AI agent that can do more than answer questions: it can open a browser, fill out forms, negotiate, keep working after a user closes the app, and ask for approval before sensitive actions such as sending email or making a purchase. The service is rolling out in the U.S. on iOS, Android, muse.ai and WhatsApp, with a free tier and paid plans for heavier use; hands-on coverage described Power at $20 per month and Maximum at $100 per month. Meta says Muse runs inside Muse Secure VM, a dedicated cloud computer with its own browser, and uses a separate Sentinel agent to approve outbound internet actions before they leave the virtual machine. Amazon cut off Muse after saying Meta had not notified it that Muse would access the store, that the agent did not identify itself while browsing, and that it appeared to capture and store customer credentials. Amazon also said applications that buy on behalf of customers should operate openly and respect whether service providers want to participate. Meta’s launch materials say Muse cannot see people’s passwords or payment methods because credentials go into secure storage, and that users choose which apps are connected, can revoke access, can opt out of model-training use, and receive an audit trail for what the agent has done and plans to do. The clash matters because it moves AI agents from demos into the economics of retail distribution. Amazon generated more than $68 billion in ad revenue last year, a business tied to product discovery and sponsored placements on its own pages, while Meta is trying to make Muse a mainstream assistant for chores such as shopping, travel, email, schedules and long-running goals. The two companies are also business partners: Amazon products have been purchasable inside Facebook and Instagram since 2023, and Meta has signed a multibillion-dollar deal to run agentic AI workloads on Amazon’s Graviton chips. Muse’s early momentum has been notable. A week after launch, it was reported as the No. 1 free app in Apple’s U.S. App Store, ahead of ChatGPT, and early users have described practical uses such as switching an auto insurance policy, finding discount codes at checkout and building an online grocery cart. In practical testing, Muse’s delegated shopping and email handling were described as genuinely useful but uncomfortable because the system becomes more helpful as users grant more account access. That tension is now the product’s central limitation: even if a user wants an agent to act, a merchant may still reject the agent, and high-value tasks require trust in permission controls, credential handling and model reliability. Meta has acknowledged that prompt injection remains an open problem, that Muse will sometimes make mistakes, and that its stronger Muse Confidential VM mode is planned for later this year rather than fully available now. The company has also opened a bug bounty program offering up to $300,000 for valid reports, including successful prompt-injection attacks. For the agent market, the Amazon block is a practical warning: consumer demand, app-store momentum and secure-agent architecture may not be enough unless major platforms agree on identification, opt-outs, liability and who owns the customer relationship when software shops on a person’s behalf.