Meta's AI Assistant Muse Expands to Mac as Amazon Blocks It From Shopping Platform

Meta's AI assistant Muse launched on Mac this week, after previously rolling out on mobile and web, and briefly topped the U.S. App Store rankings. The key change in this Mac version isn't the interface but the permissions: Muse can directly interact with files, messages, calendar, notes, and email within native apps. This means it is no longer just a chat window, but is allowed into the core data layer of users' daily work.

But the expansion of permissions is conditional. As on other platforms, users control Muse's access permissions on an opt-in basis, and the app requests approval before performing sensitive operations. Mark Zuckerberg mentioned the fast pace of development when announcing the Mac version's release. The speed at which this product line advances, and the scope it is allowed to touch, form the two premises for understanding the subsequent conflict.

Almost simultaneously, Amazon has blocked Muse from shopping on its platform, displaying an error message stating that unauthorized AI agent access violates its customer terms of use. This is the currently confirmable fact: the interception has occurred, and the reason is explicitly stated as a terms issue.

But behind the terms lies a structural background. Amazon operates its own foundation models and inference infrastructure, putting it in a direct competitive relationship with Meta on AI capabilities. The incident occurred against the backdrop of escalating tensions among tech companies around agentic commerce. Therefore, this interception can be read either as a platform enforcing access rules or as ecosystem defense against a potential competitor—under the available information, neither interpretation can be independently falsified.

Beyond competitive factors, Amazon may have a practical concern: if an AI agent makes a wrong purchase, it could cause trouble for customers and merchants, so Amazon may choose to hold off on more broadly opening its platform to AI-driven shopping. This judgment is an inference, not a confirmed motive, but it points to a genuinely existing technical constraint.

The report states that Muse's hallucination rate is relatively low, but AI models are still prone to errors. Here two levels need to be distinguished: Muse's "relatively low hallucination rate" is itself a source claim, lacking specific data or third-party verification; while "AI models are still prone to errors" is the broader industry reality. In shopping scenarios, an error is not a single irrelevant answer, but a real order, a return, a dispute between merchant and consumer. Platforms' tolerance for such risks is naturally lower than their tolerance for chat errors.

Muse's expansion is not an isolated event. Both Muse and Instinct launched voice calling features this week, with companies racing to capture consumer market share. Voice is a key step for consumer agents moving from "tool" to "always-on assistant," and also a signal of companies seizing user mindshare within the same time window.

There are two reference points in the competitive landscape: competitor Instinct is reportedly raising funds at a $10 billion valuation, a figure that is only reported and not confirmed by first parties; Poke was acquired by Cognition earlier this year. One is raising funds, one was acquired, showing that this track is both attracting capital and undergoing consolidation. Many in the industry believe consumer AI agents are the core of computing's future, set to disrupt the traditional app ecosystem and software business—this is an industry expectation, not a result that has already materialized.

Putting the two lines together, the tension becomes clear: what Muse gains on Mac is deep access within the user's device, while what Amazon shuts is the cross-platform transaction channel. The former is driven by user authorization, the latter by platform rules. When an AI agent tries to cross from "help me read" to "help me buy," it enters not the user's own data space, but the territory of another commercial entity.

It is currently unclear what specific technical means Amazon used to block Muse's shopping and how long it will last, nor is it clear whether and when Amazon will more broadly open its platform to AI-driven shopping. The industry rules or regulatory direction related to agentic commerce are likewise unclear. What is certain is that this interception has drawn a real boundary for all consumer AI agents: capabilities can iterate quickly, but the speed at which transaction permissions open depends on the platform, not the model.