---
title: "Should you block AI agents from your website? | PostCliq"
description: "Amazon blocks AI agents and Walmart built them a gateway. Should your website block AI agents? The data, why most blocks are accidental, and a simple policy."
url: "https://postcliq.com/blog/block-ai-agents"
---

# Should you block AI agents? Amazon says yes, Walmart says come in

Amazon blocked Meta's Muse. Walmart opened a gateway for agents with house rules. Most websites block agents by accident. Let agents read and enquire, and keep payments human.

By [Nirmal Gyanwali](https://postcliq.com/blog/author/nirmal-gyanwali) Updated 23 October 2026 8 min read

On this page

1. The short answer
2. What did Amazon and Walmart actually do?
3. How common is blocking AI agents?
4. Why do websites block AI agents by accident?
5. Should you block AI agents from your website?
6. Where is agent access heading?
7. How to check whether your site blocks AI agents
8. What to do next
9. Frequently asked questions

## The short answer

Key takeaways

- Amazon began blocking Meta's Muse shopping agent on 20 September 2026; JPMorgan Chase confirmed on 13 October that it blocks outside agents from logging in to customer accounts.
- Walmart launched its Agent Gateway on 21 October 2026: a Model Context Protocol (MCP) server for approved agents, with spending limits and per-agent profiles to follow.
- Jefferies tested several hundred large US shopping and travel sites: about a third blocked Muse and another third challenged or restricted it.
- Many blocks are accidental. Walmart said Muse's checkout failures on its own site were unintentional, caused by a human-verification step.
- Our view: let agents read and enquire freely, and put human confirmation on payments and account changes. Block the risky action, not the customer's agent.

The question used to be about crawlers: do you let AI companies read your pages to train models or answer questions? That debate is not over, and our guide to [ClaudeBot and Anthropic's other bots](https://postcliq.com/blog/claudebot) covers it. The new question is sharper. A person has asked an assistant to find a plumber, book a table or buy a pair of shoes, and the assistant has arrived at your website. Do you let it in?

## What did Amazon and Walmart actually do?

Amazon chose control. Walmart chose a managed front door. Amazon began showing Muse users a notice in September 2026 that access by an unauthorised AI agent breached its conditions of use. Walmart announced on 21 October that approved agents can search and buy through a dedicated gateway, under published rules, with the shopper in charge of spending.

Amazon's stated reasons were specific. It said Muse did not identify itself as an agent, that Amazon had not agreed to take part, and that the way Muse handled customer credentials raised privacy and security concerns, as [CNBC](https://www.cnbc.com/2026/10/13/jpmorgan-ai-agents-account-access.html) reported. Amazon cut access after asking Meta to leave Amazon out of the product, which Meta did not do.

Walmart's announcement, [published on its corporate newsroom](https://corporate.walmart.com/news/2026/10/Making-It-Possible-for-Agents-to-Shop-on-Behalf-of-Customers) , has three parts.

- **Agent Gateway.** An MCP server, launched as a beta for companies and developers, that lets approved agents search Walmart and Sam's Club inventory and buy for the shopper. Walmart describes it as faster and more reliable than an agent clicking through the website as a person would.
- **Agent Profiles.** Due within a month. Each agent gets its own profile in the customer's account, so its purchases show separately. The shopper can approve every purchase or set a spending limit, and can switch an agent off at any time.
- **Terms for agents.** Agents never see card details and cannot change payment methods. They may use only the information the task needs. Walmart's returns process applies even when an agent gets it wrong.

> Agents that do not follow these terms will be blocked from Walmart's site.

Walmart, 21 October 2026

Note the last line. Walmart is not saying yes to every agent. It is saying yes to agents that identify themselves and follow rules, and no to the rest. That is a policy, not a reflex, and it is the more useful model for everyone else.

## How common is blocking AI agents?

More common than most site owners realise, and growing. Jefferies tested several hundred leading US e-commerce, airline, travel and hotel websites with Muse. About a third blocked it, about a third challenged or restricted it, and only about a third were easy for it to use. Jefferies said new restrictions were appearing almost daily ( [Proactive](https://www.proactiveinvestors.com/companies/news/1099848/meta-s-muse-ai-agent-blocked-by-one-third-of-websites-tested-jefferies-says-1099848.html) , 9 October 2026).

Agents at the door, October 2026

**1 in 3** large US shopping and travel sites tested blocked Meta's Muse, and another third restricted it [Jefferies, via Proactive, Oct 2026](https://www.proactiveinvestors.com/companies/news/1099848/meta-s-muse-ai-agent-blocked-by-one-third-of-websites-tested-jefferies-says-1099848.html)

**Over half** of the customer environments Cequence studied saw Muse traffic within about three weeks [Cequence, via SecurityBrief, Oct 2026](https://securitybrief.com.au/story/meta-s-muse-ai-agent-seen-in-half-of-customer-traffic)

**Nearly 6x** growth in Muse traffic at Cequence's median customer within about two weeks of first appearing [Cequence, via SecurityBrief, Oct 2026](https://securitybrief.com.au/story/meta-s-muse-ai-agent-seen-in-half-of-customer-traffic)

Two caveats on those numbers. Jefferies tested large US sites, not small Australian ones, and its sample was not published. Cequence sells agent-control software and did not publish a customer count. Both point the same way all the same: agents are arriving quickly, and a large share of websites are turning them away.

The blockers are not only retailers. [TechCrunch](https://techcrunch.com/2026/10/06/the-next-hurdle-for-ai-agents-getting-websites-to-let-them-in/) found Delta and United restricting automated access, Yelp barring non-human traffic unless the agent pays for its data licensing programme, and Zillow's anti-scraping defences sometimes catching agents it says it allows. JPMorgan Chase confirmed it blocks outside agents from logging in with customers' banking credentials to act for them, while read-only access such as checking a balance through a data connector still works ( [CNBC](https://www.cnbc.com/2026/10/13/jpmorgan-ai-agents-account-access.html) ).

## Why do websites block AI agents by accident?

Because the tools that stop bad bots cannot easily tell a customer's agent from a scraper. A CAPTCHA, a human-verification button, an aggressive firewall rule or a form that only works with a mouse will stop an agent as surely as a deliberate ban. Nobody decided to turn the customer away. The settings did.

Walmart is the clearest example. It is a Muse launch partner, yet Muse's checkouts were failing on its site. Walmart told TechCrunch the failures were not intentional: a human-verification button interrupted the agent and the check failed. If a retailer that wanted agents in still blocked them, a small business that has never thought about agents almost certainly has a setting doing the same.

The identification problem makes it worse. Cequence reported that Muse presented as a standard Chrome browser, ran from a cloud browser through a consumer VPN, and did not identify itself as an agent ( [SecurityBrief](https://securitybrief.com.au/story/meta-s-muse-ai-agent-seen-in-half-of-customer-traffic) , 22 October 2026). You cannot write a clean rule for a visitor you cannot recognise, so bot defences fall back on behaviour, and an agent filling in a form at machine speed looks suspicious.

For a business whose website is its shop window, this is the PostCliq problem in a new form. An enquiry is lost after the click, and nobody sees it happen. The firewall logs a challenged request, if anyone reads the log. The form records no submission, because the agent never got that far. The customer simply buys from someone else.

## Should you block AI agents from your website?

For most businesses, no. Block abusive behaviour, not agents as a class. Let agents read your pages and use your enquiry or booking path, and require a human to confirm anything that moves money or changes an account. That is the position Walmart took, and it matches the advice the security vendors themselves are now giving.

Hari Nair, Cequence's vice president of product management, put it in one line: "block the action, not the agent" (via [SecurityBrief](https://securitybrief.com.au/story/meta-s-muse-ai-agent-seen-in-half-of-customer-traffic) ). Even Jefferies, an investment bank with no stake in your enquiry form, called blanket blocking "the wrong call over the long term".

Our view: Amazon can afford to block because Amazon is the destination. People will open Amazon themselves if their assistant cannot. A local accountant, a physio clinic or a mid-sized manufacturer has no such pull. If an assistant cannot read your prices or submit your form, it will move to the next supplier on its list, and the customer behind it will never know you were considered. Turning away a personal agent increasingly means turning away the customer who sent it.

| What the agent wants to do | Our recommendation | How |
| --- | --- | --- |
| Read pages, prices, hours and services | ✓ Allow | Server-rendered HTML; no CAPTCHA on content pages; bot rules that challenge abuse, not all automation |
| Send an enquiry or request a booking | ✓ Allow, with confirmation | A form that works without JavaScript tricks; a confirmation email the person must see |
| Pay, change an account or access personal data | ◐ Allow only through an approved path | Human approval for each payment, or a spending limit, as Walmart is doing; never share stored credentials with an unknown agent |
| Scrape the catalogue, ignore rate limits, hammer the site | ✕ Block | Rate limiting, firewall rules on behaviour, and a robots.txt that says what you allow |

Source: PostCliq analysis, drawing on Walmart's agent terms and Cloudflare on verified bots

## Where is agent access heading?

Towards identity and permission, not open doors or locked ones. The pieces are being built now. Agents will increasingly prove who they are, show whose errand they are running, and ask for the access they need. Websites that can read those signals will let good agents in and keep bad ones out without guessing.

Three developments from the past month show the direction.

- **Signed agents.** Cloudflare already recognises agents that cryptographically sign their requests, a scheme called Web Bot Auth, and lists them as [signed agents](https://blog.cloudflare.com/signed-agents/) . On 16 October it said it will also score verified bots on whether they keep honouring robots.txt and slow down when told to, publishing trust bands on Radar and letting customers pick a threshold. That is internal for now, with no dates ( [Cloudflare](https://blog.cloudflare.com/verified-bots-accountability/) ).
- **A permission protocol.** Sierra and Meta published a draft Personal Agent Protocol, nicknamed Poppy, on 9 October. The customer decides what access their agent gets, and the business sets what agents may do, through its website, its APIs or its own agent. Walmart, Shopify, Stripe and OpenAI are among the partners. It is a draft, and nobody has shipped it to consumers yet ( [Sierra](https://sierra.ai/blog/poppy) ).
- **Dedicated agent doors.** Walmart's gateway follows a path Shopify took first. Our article on [Shopify agentic commerce](https://postcliq.com/blog/shopify-agentic-commerce) shows how every Shopify store now publishes instructions telling agents how to shop it.

Our view: within a couple of years, "do you allow agents?" will sound as odd as "do you allow mobile phones?". The real question will be which agents, doing what, with whose permission. Small businesses will not build their own gateways. They will inherit them from their platform, their CDN or their booking system, and the job will be to switch them on and keep the website readable underneath.

These agents are mostly US-first today. Muse launched for US users ( [TechCrunch](https://techcrunch.com/2026/09/08/meta-debuts-its-muse-ai-agent-will-consumers-trust-it/) ), and auto browse for Gemini in Chrome, previously limited to North America, reached India only this month ( [PiunikaWeb](https://piunikaweb.com/2026/10/07/google-gemini-in-chrome-android-auto-browse-india/) ). The assistants arriving in Australia over the next year, described in our guide to [AI personal assistants and your website](https://postcliq.com/blog/ai-personal-assistants-websites) , will meet whatever settings Australian sites have by then.

## How to check whether your site blocks AI agents

Start with the free checks, then test the path a customer's agent would take. Most blocks sit in four places: robots.txt, the firewall or CDN, a CAPTCHA or verification step, and pages that need JavaScript to show their text.

On WordPress sites, the block usually lives in a security plugin, a firewall rule or a CAPTCHA added to the contact form after a spam wave, and it is rarely revisited. Separating the rules that stop attacks from the rules that stop customers is careful work, and a WordPress specialist such as WP Creative is the right person to do it without weakening the site's defences. [WP Creative's guide to securing a WordPress site](https://wpcreative.com.au/wordpress-security/) is a good place to see what each layer protects before you loosen any of it.

## What to do next

1. **Run the** [**AI crawler access checker**](https://postcliq.com/free-tools/ai-crawler-access-checker) to see which AI bots your robots.txt and firewall let in.
2. **Run the** [**agent readiness check**](https://postcliq.com/free-tools/can-ai-agents-read-my-site) to see whether an agent can read your pages without running scripts.
3. **Run** [**can visitors contact you**](https://postcliq.com/free-tools/can-visitors-contact-you) , then submit your own enquiry form from a cloud browser. If a CAPTCHA appears before the form loads, an agent is likely stopped there.
4. **Write down your policy** using the three levels above: read, enquire, pay. Share it with whoever manages your hosting, CDN and security plugin.
5. **Keep payments human.** Do not let any unknown agent store or reuse customer credentials. Use confirmation steps instead of blanket blocks.
6. **Publish a clear summary for agents** , such as an [llms.txt file](https://postcliq.com/blog/llms-txt) , once the door is open. A signpost is useless if the gate is shut.

## Frequently asked questions

### Should I block AI agents from my website?

For most businesses, no. Block abusive behaviour such as scraping and ignoring rate limits, but let agents read your pages and use your enquiry path. Require a human to confirm payments and account changes. Blocking agents as a class turns away the customers who sent them.

### Why did Amazon block Meta's Muse?

Amazon said Muse did not identify itself as an AI agent, that Amazon had not agreed to take part, and that the way Muse handled customer credentials raised privacy and security concerns. Amazon started blocking it on 20 September 2026, after asking Meta to exclude Amazon from Muse.

### What is Walmart's Agent Gateway?

It is a Model Context Protocol server, launched on 21 October 2026 as a beta for companies and developers, that lets approved AI agents search Walmart and Sam's Club and buy for shoppers. Agent Profiles, due within a month, let shoppers approve each purchase or set a spending limit per agent.

### How do I know if my website blocks AI agents?

Check robots.txt, your firewall or CDN bot settings, any CAPTCHA or verification step, and whether your pages show their text without JavaScript. Free tools can check access and readability in a minute. Then submit your own enquiry form from a cloud browser to see where an agent would stop.

## Is your site turning agents away? Check it free.

PostCliq checks crawler and firewall access, readable HTML and the enquiry path, and puts three fixes first.

Free. No credit card.

![Nirmal Gyanwali, founder and chief executive of WP Creative.](https://postcliq.com/authors/nirmal-gyanwali.jpg)

Written by

[Nirmal Gyanwali](https://postcliq.com/blog/author/nirmal-gyanwali)

Founder and CEO, WP Creative · Sydney, Australia

Nirmal Gyanwali has spent more than twenty years building websites, and most of that time watching them fail at the one job the business actually bought them for, which is turning a visitor into an enquiry. He founded WP Creative in Sydney in January 2016.

- [LinkedIn](https://www.linkedin.com/in/nirmalgyanwali/)
- [YouTube](https://www.youtube.com/@nirmalgyanwali)
- [Instagram](https://www.instagram.com/nirmalgyanwali)
- [TikTok](https://www.tiktok.com/@nirmalwpc)

## Keep reading

- [**AI personal assistants: what Instinct, Dots, Muse and Spark mean for your website**](https://postcliq.com/blog/ai-personal-assistants-websites)
- [**Shopify agentic commerce: what Shopify is building for AI agents, and what your website needs**](https://postcliq.com/blog/shopify-agentic-commerce)
- [**ClaudeBot: what it is, what it reads, and whether to block it**](https://postcliq.com/blog/claudebot)

[More on AI search and visibility](https://postcliq.com/blog/topic/ai-search) · [All articles](https://postcliq.com/blog)