An assistant’s ability to use a website depends on the website’s own access rules and account protections.
The Verge reports that in recent weeks users of Meta’s Muse, xAI’s Grok, OpenAI’s Dots and Instinct have reported blocks stopping their AI agents from doing tasks on top websites. Yelp told TechCrunch it only allows AI agents that pay for its data licensing program, and Delta Air Lines is preventing them from booking tickets on behalf of users. JPMorgan Chase and Charles Schwab have confirmed they are blocking AI agents from accessing accounts over security concerns. Some users say sites that used to work, including Cars.com and Amtrak, are no longer accessible, which suggests other companies could be blocking agents quietly.
The article notes Amazon was already fighting agents last year, when it banned Perplexity’s agentic shopping feature and sued the startup. The page presents this as a tension between companies promising agents that “do anything” and the sites those agents need. The Cars.com and Amtrak cases are user reports, not confirmations from the sites.
In plain English
An “AI agent” is software that acts for you, such as booking or buying. “Data licensing” means paying a site to use its content. “Blocking” here means a site refusing automated visitors.
Why it matters
If shoppers hand tasks to agents, a brand that blocks them may be invisible to those buyers, and a brand that allows them gains reach but loses control of the experience. Decide which pages and actions you will allow, on purpose.
An agent's ability to click through a website does not guarantee that the website will permit the activity. The service controls its own access rules, account protections and commercial relationships. Those constraints become part of the practical boundary of an assistant promising to complete tasks across the web.
For a user, the distinction between researching a purchase and completing it is particularly important. Public information may be available even when account access or a booking action is blocked. A product demonstration should make clear which stage it can reliably complete and where a person will need to take over.
For website operators, the choice need not be described as allowing everything or blocking everything. Reading public pages, collecting licensed data and taking actions within a signed-in account are different activities. A useful policy describes those activities explicitly and provides a support path when legitimate users encounter restrictions.
The reported blocks also complicate measurement. A failed agent visit does not necessarily mean the customer lost interest, and an allowed visit does not necessarily represent a completed sale. Companies evaluating this channel need to understand the full journey. The named examples show a friction point in agent adoption, but user reports about individual sites should not be treated as confirmed changes to every site's policy.