AI agents are gradually entering a phase in which, instead of offering suggestions or answering users' questions, they can act on their behalf directly on the internet. However, it seems not all major platforms are ready to accept such users.
Users of Meta's AI assistant Muse encountered a message on Sunday night while attempting to purchase items from Amazon indicating that this AI agent was not allowed to continue operating on the store.
The displayed error message read: “Continued access by an unauthorized AI agent violates Amazon’s Terms of Use that our customers have agreed to.”
In practice, this message means Muse cannot make purchases on Amazon like a regular customer. Therefore, a user who intends to complete the purchase via Meta’s AI agent must find the desired product on another platform or complete the purchase personally to finalize the order.
At first glance, this event can be seen as part of a broader competition among tech giants. Amazon itself has heavily invested in AI and possesses a suite of foundational models and processing infrastructure. In such circumstances, allowing a competitor’s AI agent to operate freely in one of Amazon’s most important business areas-e-commerce-is not necessarily a simple decision.
However, the issue is not only competition between two companies. The entry of AI agents into online shopping raises important questions about responsibility for orders and the interaction among the customer, the AI agent, and the seller.
For example, if Muse orders the wrong product, selects an incorrect quantity, or for any other reason makes an inappropriate purchase, who will ultimately be responsible for addressing the problem? In such a situation Amazon may have to deal both with the dissatisfied customer and the seller whose order was affected-even though the initial decision was made by an AI system owned by another company.
This issue is particularly important because even the most advanced AI models are not completely free of errors. Muse is among the models considered to have a low hallucination rate, but that rate is still not zero. A small mistake in the shopping environment can lead to placing incorrect orders, selecting unsuitable products, or creating disputes between buyer and seller.
GeekWire was the first outlet to observe and report this restriction. Amazon’s action suggests that e-commerce companies may require more controlled standards and mechanisms to accept AI agents-especially when these systems are intended to make commercial decisions without direct human intervention.
According to TechCrunch, on the other hand, this development does not necessarily mean Amazon opposes the concept of “agent-driven commerce.” The company may itself take the economic opportunities of such technology seriously, but prefer that a few more generations of the technology be developed before providing broad access to AI agents so that potential errors and liabilities can be better managed.
If AI agents become widespread, such restrictions could become one of the major issues in online commerce; stores will need to specify whether they accept an AI agent like a real customer or apply different rules and access for it.