AI agents hit website blocks as Meta and Walmart work on a standard

When a human check stops the agent

Some blocks are deliberate. Others come from familiar anti-spam measures that were built for a web used by people and conventional bots. A site’s “confirm you’re human” check can fail mid-process, leaving an agent unable to continue.

TechCrunch saw complaints that Muse could not complete purchases on Walmart. NBC reported that Walmart encountered the problem while testing the agent. Walmart said the blocks were not intentional and that it is working with Muse, a partnership announced at Meta Connect in September. The retailer says it wants to be where its customers are—including inside AI-agent services.

The distinction matters to users: a failed task can make the agent look unreliable even when the site caused the failure, or make a business seem hostile when its defenses were not aimed at that agent.

Meta, Walmart, Stripe, Sierra, Genesys, Rocket, NiCE and Decagon are developing an open standard for interactions between agents and businesses. Meta says the protocol will focus on agents communicating with each other for online shopping, and on distinguishing legitimate agents acting for users from malicious bots.

Access depends on the company

Airline bookings are a prominent proposed use for agents, but users say airlines reject their requests. Delta told TechCrunch it takes steps to protect customers from unauthorized automated activity. The airline has no partnership or integration that lets third-party agents search for and book tickets on its digital platforms for a customer, said Hina Chawda, Delta’s managing director of global communications. Delta is evaluating how technologies including external agents could improve the passenger experience.

United pointed to its website terms, which prohibit using automated tools or manual methods to monitor or copy the site, or for other unauthorized purposes, without prior written permission. It did not answer when asked whether it blocks particular personal agents, such as Muse.

Users have also reported problems with Yelp, eBay, Zillow, Pizza Hut, Adidas and airlines. Two said eBay suspended their accounts because they used an AI agent; another said Google disabled Instinct while it was cleaning up their Gmail inbox.

Cloudflare sits on another part of the access path. It protects websites from AI agents that may train on their data and helps manage other automated traffic. Some have linked Muse’s problems to Cloudflare’s September 15 change to crawler defaults, which lets site owners block AI bots collecting training data while allowing other bot traffic. The change also meant some sites already configured to block AI bots for security reasons began blocking AI agents on advertising pages.

Cloudflare said it had no specific data on personal agents such as Muse and pointed to its public Cloudflare Radar service. Radar shows rising bot activity, but does not distinguish legitimate bots from malicious ones. Cloudflare also has a business interest in limiting AI traffic: it launched a marketplace where AI bots are expected to pay for data, and recently began testing a browser designed specifically for AI agents.

The missing piece is predictable permission

I think the standard matters less as a promise of universal access than as a way to make permission legible. Meta’s brand partnerships could tell Muse users which businesses allow the agent in, while giving them a clearer way to report access failures. But the standard is still being developed, and many recently announced partners are not yet in Muse’s connector list.

Meta argues that turning away a personal agent is equivalent to turning away the customer. That is a useful argument for a company trying to expand its agent, but it does not settle the harder question: who gets to decide what an agent may do, and on what terms? Until businesses can set those terms clearly—and agents can reliably respect them—a failed checkout will keep looking like a product failure, even when the real problem is that the web has no shared way to say yes.