Amazon Blocks Meta’s Muse

Amazon has blocked Meta’s Muse AI shopping agent from accessing its website, citing a breach of its Conditions of Use.

The dispute highlights a growing issue for websites as AI agents increasingly browse online shops and carry out tasks on behalf of users. It also raises questions about how website owners can identify and restrict automated AI activity when traditional tools such as robots.txt may not apply.

The block was reported on 20 September 2026, after people using Muse to shop on Amazon were shown a message stating that continued access by an unauthorised AI agent violated Amazon’s Conditions of Use.

Amazon confirmed that it had blocked the service.

Amazon Relies On Its Conditions Of Use

Amazon said Meta had not informed the company that Muse would access its website and claimed that the agent does not identify itself while browsing.

The retailer also raised concerns that Muse appeared to capture and store customer credentials, which Amazon said could create privacy and security risks.

Amazon has separately asked Meta to remove Amazon from the Muse shopping experience.

The Conditions of Use are an agreement between Amazon and its customers, rather than an agreement directly between Amazon and Meta. This makes the dispute particularly interesting because the person using Muse has already accepted Amazon’s terms when using the site.

Robots.txt Could Not Specifically Block Muse

One of the traditional ways websites control automated access is through a robots.txt file.

This file can contain instructions for particular user agents, which are identifiers sent by software when requesting pages from a website.

However, Meta’s published crawler documentation does not list a specific user agent for Muse. This means Amazon could not simply add a dedicated Muse rule to its robots.txt file.

Amazon’s robots.txt does contain restrictions for a number of Meta’s other crawlers.

These include meta-externalagent, used for model training and indexing, meta-externalfetcher, which retrieves individual links when requested by users, and meta-webindexer, which supports Meta AI search.

All three are instructed not to access Amazon’s pages.

Other Meta crawlers are not specifically named and therefore fall under Amazon’s general rules for unidentified crawlers. Those rules restrict access to areas such as shopping carts, sign-in pages and account sections while leaving some product pages accessible.

Muse Browses As The User

Meta has previously explained that Muse is designed to browse the internet in a way that appears as the user’s activity.

The AI agent operates through a browser based on Chromium and uses a dedicated virtual machine for each user.

Meta says users share their own dedicated computer in the cloud with Muse. The environment is isolated and can be accessed through Muse’s apps, WhatsApp and its website.

This approach differs from some AI tools that run directly on a user’s own computer.

The distinction could become important when determining who is technically accessing a website and where the automated activity originates.

Amazon Raises Questions About Credentials

Amazon also claimed that Muse appears to capture and store customer login credentials.

Meta’s own documentation provides some detail about how authentication is handled. According to the company, credentials are stored within the user’s virtual machine rather than centralised Meta infrastructure.

Muse itself is said to operate using surrogate tokens rather than directly accessing the user’s actual credentials.

Meta says a separate component called Sentinel replaces those tokens with the real credentials at the network boundary when necessary.

This means the AI model itself is not supposed to see the user’s passwords.

However, the available reporting had not established an independent verification of Meta’s claims about how the credentials are handled.

Amazon had also not publicly provided detailed evidence explaining the basis for its concerns about credential storage.

Amazon Has Its Own AI Shopping Tools

The dispute is also taking place as Amazon develops its own AI-powered shopping services.

Amazon launched Alexa for Shopping in May 2026 and also operates another shopping agent called Buy for Me.

Amazon has said Buy for Me identifies itself when accessing websites and allows brands to opt out.

That provides some context for Amazon’s position that third-party shopping agents should be transparent and respect decisions made by individual retailers about whether they want to allow automated access.

At the same time, Amazon operates a major advertising business that depends heavily on shoppers visiting its website and viewing products and sponsored listings.

The Perplexity Case

Amazon’s dispute with Meta follows a separate legal battle involving AI company Perplexity.

Amazon previously took legal action over Comet, Perplexity’s AI-powered browser, which can log into a user’s Amazon account and make purchases on their behalf.

In March 2026, Amazon obtained a court order restricting Comet from accessing password-protected areas of its website while the legal dispute continued.

The Ninth Circuit Court of Appeals later overturned that order on 4 August 2026.

A key part of the ruling concerned who was actually accessing Amazon’s systems. Because Comet operated on the shopper’s own computer, the court found that the shopper was the person accessing Amazon rather than Perplexity’s servers.

The court did not completely rule out the possibility of a different outcome where an AI company’s own infrastructure controls the agent and accesses a website directly.

That distinction could be relevant to Muse because Meta operates the cloud computer on which its agent runs.

What This Means For AI Agents

The Amazon and Meta dispute illustrates some of the difficulties website owners may face as AI agents become more capable of browsing and purchasing online.

Traditional controls such as robots.txt depend on websites being able to identify a crawler through its user agent. An AI service that does not provide a dedicated identifier can make that approach less straightforward.

Website owners can also rely on terms of service, authentication controls and server-level measures, but the legal position around AI agents continues to develop.

For now, Amazon has chosen to rely on its Conditions of Use to restrict Muse.

The wider question is whether AI shopping agents should identify themselves clearly when browsing websites and what rights businesses have to prevent them from accessing their services.

As more AI tools begin acting on behalf of consumers, these issues are likely to become increasingly important for both retailers and website owners.