AI agents are moving from merely helping users search to actually placing orders on their behalf, but whether e-commerce platforms have the right to bar these agents from their transaction systems is becoming the latest flashpoint among tech giants.
On September 21, GeekWire reported that Amazon has severed Meta Platforms, Inc.'s personal AI agent, Muse, from making purchases on Amazon for users. When users attempt to shop on Amazon through Muse, they see a notice stating that continued access by unauthorized AI agents violates the platform's terms of service. Amazon noted it had previously asked Meta to voluntarily exclude Amazon from Muse's service scope but received no response.
The conflict is not between two companies with no existing relationship. Since 2023, Amazon products have been directly purchasable on Facebook and Instagram. In April this year, Meta signed a multi-billion-dollar agreement with Amazon to run agentic AI workloads on Amazon's cloud infrastructure. Amazon says it is currently communicating directly with Meta, but declined to comment on whether it is considering legal action.
Why did Amazon block Muse?
Amazon's stated rationale centers on authorization, identity disclosure, and account security.
Amazon claims Meta did not notify it in advance that Muse would access the platform, and that Muse does not proactively reveal its AI agent identity during browsing. Amazon argues this amounts to an unauthorized third party entering user accounts, processing transactions, and touching related data.
Amazon also pointed out that Muse can access account pages and order history at the user's prompt. An Amazon spokesperson said third-party apps offering concierge shopping services should operate transparently and respect whether a platform permits their participation.
Meta, for its part, previously stated that Muse cannot access user passwords or payment methods. Credentials provided by users are placed in secure storage, and Muse can use the services without directly viewing those credentials.
Muse's product design also sits squarely at the center of this dispute: for services offering public APIs, Muse can connect using user credentials; if no API exists, it can "use the service through a browser just like a user would." The access method Amazon has blocked is precisely the latter.
From lawsuits to terms of service
Amazon has previously tried to restrict AI agent access to its platform through legal means, but the outcomes of those cases have not been entirely favorable.
Amazon previously filed a lawsuit over Perplexity's Comet browser and obtained a preliminary injunction in March this year. However, on August 4, the Ninth Circuit Court of Appeals overturned that injunction, ruling that under federal anti-hacking law, it is the user, not the AI company, who accesses Amazon's computer systems. On September 10, the court further denied Amazon's petition for rehearing.
Still, that ruling leaves room for Amazon to pursue claims based on contracts and terms of service. The notice targeting Muse does not invoke hacking behavior either; it points directly to Amazon's "terms of use," indicating the core of the dispute is shifting to whether AI agents are authorized by the platform.
AI purchasing agents are touching the core interests of e-commerce platforms
Amazon's hardline stance on agent-assisted shopping is also tied to its own business model.
Amazon's advertising revenue in 2025 exceeded 68 billion U.S. dollars. If AI agents handle search, price comparison, and order placement directly, users would no longer need to enter Amazon's pages, potentially undermining the traditional ad display and product recommendation mechanisms. Over the past year, Amazon has continuously restricted external AI agents from accessing its platform. In addition to suing Perplexity, the company has also moved to block shopping agents from Google and OpenAI.
Meanwhile, Amazon is developing its own agentic shopping capabilities. In the "Alexa for Shopping" feature launched in May this year, "Buy for Me" can go to external brand websites on the user's behalf to find and purchase items. Amazon emphasizes that this feature proactively discloses its identity and allows brands to opt out.
From this, a core question raised by AI-assisted purchasing is coming into focus: whether platforms will allow third-party AI agents to enter their transaction systems on behalf of users, and whether platforms can continue to control the transaction gateway and user relationships.
Rapid growth of Muse accelerates the battle over rules
Meta officially launched Muse on September 8, positioning it as a personal AI agent capable of performing multi-step tasks. It can connect to email, calendars, payments, dining, and shopping services, and supports iOS, Android, muse.ai, and WhatsApp.
Muse quickly gained traction after launch. Within just one week of release, it topped the free app chart on Apple's U.S. App Store, surpassing ChatGPT. Early users have already used it to switch car insurance plans, find discount codes, and auto-fill online shopping carts.
Meta says Muse runs in a secure virtual machine with a dedicated browser, and it asks for user confirmation before sensitive actions like sending emails or completing purchases. A system called "Sentinel" monitors and reviews content sent out by the agent.
As Muse moves into real-world transaction scenarios, the boundary between AI agents and traditional internet platforms is being pushed further into the spotlight. Who controls the agent, who owns the user relationship, which services agents can access, and how transaction value is distributed are all questions that will need to be resolved in the era of "agent commerce."
Comments