Independent AI agents are facing growing restrictions from prominent websites and online services, limiting their ability to perform tasks promoted by companies such as OpenAI, Meta, and Google. These tasks include booking restaurants and flights, shopping, and organizing email on behalf of users.
Users of Meta’s Muse, xAI’s Grok, OpenAI’s Dots, and Instinct agents reported that they could no longer complete tasks on websites that had previously been accessible. Yelp told TechCrunch that it only allows agents participating in its paid data licensing program, while Delta Air Lines prevents agents from booking tickets on behalf of customers. JPMorgan Chase and Charles Schwab also confirmed that they block AI agents from accounts because of security concerns.
The Problem Goes Beyond Direct Blocking
The restrictions are not limited to explicit decisions to ban agents. Anti-bot and fraud-prevention systems may treat them as suspicious programs, even when the website did not design the barrier specifically against AI. Users reported problems with Amazon, Cars.com, and Amtrak. Some accounts using Muse were also reportedly blocked on Google for “suspicious activity,” while payments on Walmart stalled despite its partnership with Meta.
Chris Canal, CEO of AI safety assessment company EquiStamp, said these obstacles effectively reduce the value of agents, because their core promise is the ability to do what a user can do online. As more websites block them, the range of tasks they can complete shrinks.
The Amazon Dispute Illustrates the Legal and Commercial Complexity
Amazon had banned Perplexity’s agentic shopping feature last year and then sued the company. Amazon said that third parties purchasing products on behalf of customers should operate transparently and respect service providers’ decisions about participation. Perplexity, by contrast, argued that Amazon rejects agents because it could lose human traffic and advertising revenue, and considered the agent an extension of the user.
A judge later ordered Perplexity to stop agentic shopping on Amazon, but an appeals court overturned the decision in August. Amazon then filed an amended complaint. The case continues to reflect a broader question: Is an agent a user acting on behalf of a person, or an external program imposing automated traffic and new commercial and security conditions on the website?
What Is Changing in Practice?
Some companies are trying to address the problem through partnerships rather than resisting agents. Meta has enabled Muse to shop at Walmart, Gap, Sephora, Best Buy, and others, while Shopify has opened its network of stores to agents. But this approach means that an agent will not automatically work across the entire web; instead, it will operate within a group of services that have previously agreed to grant it access.
Cloudflare is making unauthorized access more difficult. In 2025, it introduced a one-click option for blocking AI crawlers, then began blocking crawlers that collect data simultaneously for search-engine indexing and AI-model training. The company says that bot traffic puts pressure on internet infrastructure and affects email, ticket purchases, bookings, and product sales.
The current result is not the end of AI agents, but it reveals their reliance on an access layer that they do not fully control. In the future, users may have to undergo more tests to verify that they are human, even for everyday services such as shopping from online stores. Questions about consent, data licensing, security, advertising, and legal liability will also remain open unless websites and agent developers agree on shared operating rules.