AI agents face growing web blockade as Amazon, airlines restrict access

Personal AI agents are increasingly being blocked or disrupted by websites, creating a major obstacle for a new generation of tools designed to shop, book flights, make reservations and perform other online tasks on behalf of users.

The problem has emerged as consumer AI agents such as Meta’s Muse, Instinct and ChatGPT’s Dots move beyond answering questions to carrying out transactions directly on websites.

Amazon recently blocked Meta’s Muse from accessing its retail platform, preventing the AI agent from browsing products or making purchases.

But Amazon is not the only website creating access problems. Users have reported similar difficulties with platforms including Walmart, airlines, Yelp, eBay, Zillow, Pizza Hut and Adidas.

Some blocks appear intentional, while others are caused by conventional anti-bot systems that websites use to prevent spam, scraping and malicious activity.

The distinction is often unclear to users, leaving them unsure whether an AI agent has been deliberately rejected or simply failed a website’s existing security checks.

Walmart, for example, said it was not intentionally blocking Muse and has partnered with Meta to support the AI agent. The retailer said some users may instead be encountering its human-verification system, which requires visitors to click a button to confirm they are human.

When an AI agent cannot complete that verification, the session can be terminated.

The problem highlights a growing mismatch between traditional website security systems and an internet where AI agents increasingly act on behalf of human users.

To address the issue, Meta, Walmart, Stripe, Sierra, Genesys, Rocket, NiCE and Decagon are working on an open standard for agent-to-agent communication in online commerce.

The proposed protocol is intended to help businesses distinguish legitimate AI agents acting for customers from malicious automated traffic.

Meta said the goal was to create clearer rules that give businesses greater control while allowing consumers to use AI agents without repeatedly encountering access barriers.

“Turning away a personal agent means turning away the customer behind it,” Meta said in announcing the initiative.

Airlines have also emerged as a major challenge for agentic AI, particularly as booking flights is promoted as one of the technology’s key consumer applications.

Delta said it does not currently have a partnership that allows third-party AI agents to shop for or book flights on behalf of customers through its digital platforms.

The airline said it was evaluating how external AI agents could be incorporated while maintaining security and protecting the customer experience.

United Airlines, meanwhile, pointed to its terms of use, which prohibit users from employing robots, spiders or other automated devices to monitor or copy its website without prior permission.

Other platforms have similarly placed restrictions on automated access.

Yelp said it does not permit non-human traffic on its platform unless an agent has paid to access its content and data through its licensing programme. As a result, an AI agent attempting to obtain a quote, join a waitlist or book a restaurant table may fail without a formal partnership.

eBay said it does not prohibit all third-party shopping agents but restricts unauthorised activities such as automated scraping and model training.

Some early users have also reported that their accounts were suspended after using AI agents, while others said agents were blocked while performing tasks such as managing email.

Cloudflare, which provides security and content delivery services for websites, has also become part of the debate.

Some users and observers have speculated that changes to crawler settings by Cloudflare could be affecting traffic from personal AI agents. Cloudflare, however, said it had no specific data to share about the reported blocking of agents such as Muse and directed questions to its public Cloudflare Radar data.

The company has a growing role in managing AI-related web traffic and recently launched a marketplace through which AI bots can pay for access to data. It is also testing a web browser designed specifically for AI agents.

For Meta, direct partnerships with businesses appear to be one way of addressing the problem while broader industry standards are developed.

The company maintains a list of connectors within Muse that allow the AI agent to interact with partner services, although some newly announced partners have yet to appear on the list.

As AI agents become more capable of acting independently online, the ability of websites to recognise and safely accommodate legitimate agents is emerging as a key challenge for the next phase of consumer AI.