Why Websites Are Blocking AI Agents: The New Access Fight Between Bots and Businesses

Autonomous AI agents are being marketed as digital assistants that can search, compare options and complete online tasks. But a growing number of websites are limiting or blocking them. The conflict exposes a basic dependency of agentic AI: an assistant can only act on a service when the service permits the software to access its pages, accounts and booking flows.

What is happening

The Verge reported that users of agents from companies including Meta, xAI, OpenAI and Instinct have encountered blocks across major websites. Yelp told TechCrunch it allows agents through a paid data-licensing program, while Delta Air Lines is preventing agents from booking tickets on behalf of users. JPMorgan Chase and Charles Schwab have also confirmed that they block agents from accessing customer accounts because of security concerns.

Some difficulties may come from deliberate restrictions; others can happen when anti-bot systems mistake an automated agent for a suspicious script. The distinction matters because the user experience can look identical: the agent cannot finish the task. A block does not always mean a company has made a formal policy against every kind of AI assistance.

Why businesses set boundaries

Web services have to protect accounts, transactions and private customer data. Automated systems can make mistakes at scale, trigger fraud controls or access information outside an intended workflow. Companies also operate their own data licensing and advertising models. If an agent extracts product information or completes a purchase without using the expected interface, a business may worry about lost revenue, support burden or responsibility if something goes wrong.

AI companies often describe an agent as an extension of the user, similar to an assistant acting on a person’s behalf. Websites may see it differently: a third-party program interacting with systems that were built for a human-operated browser. Both positions raise legitimate questions about consent, access, identity, and who is responsible for an action performed by an automated intermediary.

What this means for users

Consumers should expect agent capabilities to vary by website and task. A model may be able to summarize public pages but fail when it encounters sign-in, a checkout, CAPTCHA or a booking flow. Users should not try to bypass a site’s restrictions with credential sharing or stealth automation. If the service does not support an agent, use the official app or website directly.

Before letting an agent manage purchases or accounts, review which permissions it receives, what data it can see, and whether it can submit or change information without asking. Keep approval requirements for payments, travel bookings, account changes and other consequential actions.

How the web could adapt

One possible path is explicit partnerships and licensed interfaces. The Verge notes Meta has partnered with some retailers for shopping access, and Shopify has opened its storefront network to agents. Another path is a standard that lets sites declare which actions agents may perform and how they must identify themselves. Better authentication and narrow, auditable APIs would be safer than asking an agent to mimic a person clicking through a website.

For AI developers, reliable access is becoming part of product quality. An agent that reasons well but cannot reach the service it needs is limited. For web businesses, blocking every bot may reduce risk but can also make legitimate automation harder. The durable solution will need clear rules, user consent and technical controls that distinguish helpful delegated work from scraping or abuse.

Our related articles cover the risks of AI agents accessing live systems and publisher concerns around website crawling and AI training.

Source: The Verge, October 17, 2026. The examples above reflect the reporting cited in that story; policies can differ by service and change over time.