Meta launched its Muse personal AI agent in the US on 8 September 2026, with browser-based shopping and purchase tasks among its stated uses The Verge. Reports that Muse users receive an Amazon Conditions of Use error while trying to buy items would point to a different kind of restriction from Amazon’s earlier robots.txt controls.
Amazon’s documented robots.txt action concerned automated crawlers, while Muse can use browser automation when a service has no public API Hexagon TechCrunch. The published material reviewed here does not include an Amazon statement confirming a Muse-specific enforcement action, or Meta’s account of a specific Amazon purchase failure.
What has been reported to Muse users?
Users have reported seeing an Amazon Conditions of Use error when Muse attempts to complete a purchase. The reported issue appears during a buying task rather than during a general Muse conversation.
That distinction matters. A shopping agent can search for products, open pages and collect product information without reaching the point where Amazon must accept an attempted transaction.
The available published accounts do not reproduce the full error text, identify its first appearance date, or show an Amazon response code. They also do not establish whether product search remains available in every case while product-detail pages or checkout fail.
Nor do the available reports establish whether the message appears when Muse opens a product page, adds an item to a basket, begins checkout, or submits a purchase request. Those are separate steps, and Amazon could apply different controls at each one.
Does this look like a robots.txt block?
No published evidence ties the reported Muse purchase error to robots.txt. A robots.txt file is a web directive that asks automated crawlers which pages they should not access; Hexagon reported that Amazon expanded its file in late 2025 to cover 47 AI bots Hexagon.
Hexagon said the listed restrictions covered AI crawlers from OpenAI, Google, Meta and Huawei. Its account said Meta entries were training and research bots associated with Facebook, Instagram and WhatsApp, rather than a named Muse browser session Hexagon.
That kind of crawl control is not the same as a user-facing conditions error. Robots.txt is voluntary, Hexagon noted, and large AI companies generally respect it for legal and operational reasons Hexagon.
A Conditions of Use message instead suggests enforcement after Amazon has identified activity it considers inconsistent with its rules. The available reporting does not say which rule Amazon applied, how Amazon identified Muse, or whether it acted against Meta’s software, a particular account, or a pattern of requests.
| Question | What is documented | What remains unclear | Source |
|---|---|---|---|
| Amazon crawler controls | Amazon expanded robots.txt in late 2025 to block 47 AI bots, according to Hexagon. | Whether Muse uses any bot identity covered by that file. | Hexagon |
| Muse website access | Muse can connect through a public API when one exists and can use a browser when one does not, TechCrunch reported. | Which access method Muse used on Amazon in reported failed purchase attempts. | TechCrunch |
| Reported purchase error | Users have reported an Amazon Conditions of Use message while attempting purchases through Muse. | The complete wording, trigger point, affected accounts and Amazon’s stated reason. | Not independently documented in the published material cited here. |
| Amazon confirmation | No Amazon statement about Muse appears in the available reporting. | Whether Amazon intentionally restricted Muse or applied a general automated-access control. | Not available. |
What can Muse do on a user’s behalf?
Meta presented Muse as an agent for online shopping, email and trip planning when it launched the service on 8 September 2026 The Verge. Meta said the agent could open a browser, fill in forms and continue working after a user closes the app.
Muse can ask for approval when a task requires a consequential action such as a purchase, The Verge reported The Verge. This means a shopping workflow has several stages: finding an item, inspecting it, preparing an order and seeking user approval before payment.
TechCrunch reported that Muse connects to services selected by the user, one at a time. It can use supplied credentials with a public API, or turn to a browser if a service does not offer an API TechCrunch.
That browser fallback is important for Amazon. It means Muse does not necessarily need Amazon to offer a dedicated agent integration before it can try to complete a task there.
Meta said Muse can make purchases using Link by Stripe, while Shop Pay and 1Password support were described as future additions TechCrunch. The Verge also reported that Muse had Stripe Link checkout support at launch The Verge.
Those payment options do not show how Amazon handles Muse-driven shopping. They only show that Meta designed Muse to handle the final stages of some online transactions.
How does Meta say Muse controls browser activity?
Meta says each Muse user has a dedicated virtual machine, or VM, in the cloud with a browser and storage for connected-service data Meta. Meta says credentials for connected services are stored in that user’s VM rather than in centralised Meta infrastructure.
The company says Muse itself does not see real passwords or payment credentials. Its system substitutes limited-use tokens, then inserts real credentials only after a separate component approves a concrete network request Meta.
Meta calls that separate component Sentinel. According to Meta, Sentinel is the sole authority for connector actions and outgoing network traffic, and can allow, deny or ask the user about a proposed action Meta.
Meta says Sentinel can inspect details including a request’s hostname, destination address, port, protocol, HTTP method and path before allowing it to leave the VM Meta. This describes Meta’s internal controls, not Amazon’s decision-making or enforcement methods.
For readers trying to understand the reported Amazon issue, the practical point is simple: Muse is not merely a search box sending a query to a website. It can operate a browser and make network requests through a controlled cloud environment.
What did Amazon’s earlier crawler restrictions cover?
Hexagon reported that Amazon’s late-2025 robots.txt update prevented covered bots from reading Amazon product pages, prices, specifications and customer reviews Hexagon. That would limit an assistant that depends on crawling Amazon’s catalogue to generate shopping recommendations.
Hexagon also reported that Amazon did not apply the same crawler restrictions to Zappos, Shopbop and Woot. The report characterised that difference as a controlled experiment, but Amazon did not provide a statement in the material cited here explaining its reasoning Hexagon.
The report discussed Amazon’s Rufus assistant and its Buy for Me feature. Those products do not prove Amazon’s motive for any restriction on Muse, and they should not be treated as evidence that Amazon targeted Meta to shield its own shopping tools.
The narrower conclusion is supported by the record: Amazon had already restricted a group of AI crawlers before Muse launched. That fact does not explain a later user-facing Conditions of Use message during a Muse purchase attempt.
| Access method | What it is for | What the evidence says about Amazon | Why the distinction matters |
|---|---|---|---|
| robots.txt | A voluntary instruction for web crawlers. | Hexagon reported Amazon used it to restrict 47 AI bots in late 2025. | It concerns crawling rules, not necessarily a logged-in shopping session. |
| Public API | A service’s formal software connection. | No Amazon Muse API arrangement is documented in the available reports. | Muse can use APIs where a service provides one. |
| Browser automation | Software acting through a browser when no API is available. | No Amazon explanation of its treatment of Muse browser sessions is available. | This is the route Muse can use to access services without an API. |
| Conditions enforcement | A site response based on its rules for use. | Users have reported such an error during Muse purchase attempts. | It would be distinct from a robots.txt instruction, but the trigger is unconfirmed. |
What have Meta and Amazon confirmed?
Meta confirmed Muse’s launch and described it as a personal agent that can work across connected apps and services TechCrunch. Meta also published technical documentation describing a browser-equipped VM and Sentinel approval system Meta.
Amazon has not publicly confirmed a Muse-specific block in the material available here. No Amazon statement explains whether a reported Conditions of Use message results from automated activity, account checks, checkout safeguards, crawler restrictions or another policy.
Meta has not publicly described a specific Amazon failure, including whether users can still search Amazon through Muse while product pages or checkout fail. It has also not identified the Amazon step at which a reported error appears.
What can shoppers do while the position remains unclear?
Users should treat an agent’s shopping result as a draft until they have reviewed the product, price, delivery terms and payment screen themselves. Meta says Muse can pause for approval before consequential actions, including purchases The Verge.
If Amazon displays a Conditions of Use message, users should not assume it is a temporary page error or try to bypass it through repeated automated requests. The available reporting does not establish a supported workaround from either Amazon or Meta.
Muse users can also choose which services to connect, according to TechCrunch TechCrunch. That choice matters because a connected service can give an agent access to more of a user’s routine, including email, calendars and payment-related workflows.
What could not be verified?
We could not verify the complete Amazon Conditions of Use error text reported by Muse users. We also could not verify the date it first appeared, the number of affected users, or whether it occurs consistently across search, product pages, baskets and checkout.
We could not verify whether Amazon still permits Muse to search its catalogue while limiting product-detail access or transaction completion. No public Amazon statement in the available reporting distinguishes those stages.
We could not verify whether Amazon enforced its Conditions of Use, applied an automated-risk control, relied on an account-level restriction, or used another measure. The earlier robots.txt restriction is documented, but it is not evidence that robots.txt caused the reported purchase error.
We also could not verify whether Muse used a Meta bot identity included in Amazon’s earlier crawler list. Meta’s next public explanation of the error, or an Amazon statement identifying the relevant policy and workflow step, would answer the most useful open question.
Which sources explain the available evidence?
- The Verge — Meta bets on AI agent Muse to catch up in AI race
- TechCrunch — Meta debuts its Muse AI agent. Will consumers trust it?
- CNBC — Meta pushes into personal AI agents in Muse Spark family
- NBC Miami — Meta’s AI assistant can manage your life, but expert urges caution
- Hexagon — Why Amazon Is Blocking 47 AI Bots — and What It Means for the Future of Product Discovery
- Reuters — Meta launches AI agent that can access other apps to send emails, make payments
- Meta — Security and safety for AI agents: our approach with Muse
