AI agents are being designed to shop, book flights, manage accounts, and complete everyday online tasks. But a growing number of websites are making those activities difficult or restricting them altogether, exposing a major obstacle to the future of AI-powered automation.
In an October 17, 2026, report, The Verge documented access problems affecting AI assistants from companies including OpenAI, Meta, and xAI. Some restrictions reflect explicit website policies, while others may result from security systems designed to detect automated activity.
The conflict reveals a fundamental limitation of AI agents: being technically capable of performing an online task does not guarantee permission to access the website required to complete it.
Which Websites Are Restricting AI Agents?
Several major companies have confirmed restrictions on automated access, although their policies and reasons differ.
Yelp told TechCrunch in October that it does not permit nonhuman traffic unless the agent has access through its paid data-licensing program. That makes licensing, rather than technical capability alone, a condition for permitted automated access.
According to The Verge, Delta Air Lines is preventing AI agents from booking flights on behalf of customers. JPMorgan Chase and Charles Schwab have also confirmed restrictions on unauthorized AI agents accessing customer accounts, citing security concerns.
These restrictions are particularly significant for financial services. An agent interacting with sensitive account information or initiating transactions presents different risks from one retrieving publicly available product descriptions.
Amazon has taken a particularly confrontational approach toward certain third-party shopping agents. Its dispute with Perplexity has developed into litigation over whether an AI-powered browser assistant can interact with Amazon's store on a customer's behalf despite the retailer's objections.
Meanwhile, users have reported difficulties using agents on other websites. However, an unsuccessful automated session does not necessarily prove that a company has adopted an explicit policy against AI agents. Authentication failures, fraud controls, and general anti-bot systems can produce similar results.
Why AI Agents Are Being Blocked
Traditional websites are primarily built around human visitors using browsers, accounts, and checkout interfaces. AI agents introduce a different pattern of interaction: software can navigate pages, retrieve information, select products, fill out forms, and initiate actions with varying levels of user supervision.
That behavior can resemble ordinary automated traffic, including activity that websites already attempt to limit.
Security systems may challenge or reject sessions based on unusual navigation patterns, repeated requests, unfamiliar login behavior, or other risk signals. An agent acting for a legitimate customer can encounter the same defenses intended to protect against credential abuse, spam, fraud, and unauthorized automation.
Companies also have commercial reasons to control access. Shopping platforms, travel services, and online marketplaces often depend on relationships with customers, advertising opportunities, product presentation, and control over transactions.
An outside agent that completes a purchase without the customer navigating the website could change those relationships. That possibility helps explain the commercial tension, although it does not establish that advertising or revenue protection motivates every restriction.
Another important distinction concerns AI crawlers. Crawlers collecting information for search indexing or model training are not necessarily performing the same function as an authenticated personal assistant completing a specific user-requested task.
Cloudflare, which provides web infrastructure and security services, introduced additional AI-crawler controls in July 2025, including mechanisms that give website operators more control over automated content access.
Those tools illustrate how website owners can regulate AI-related traffic. They should not, however, be interpreted as a universal prohibition on personal AI agents.
Amazon's Legal Fight With Perplexity Highlights the Conflict
Amazon's dispute with Perplexity provides one of the clearest examples of the legal questions surrounding AI-assisted web access.
Amazon filed its lawsuit in November 2025, challenging the operation of Perplexity's Comet browser assistant on its shopping platform. Amazon argued that Perplexity's automated interactions violated restrictions on access to its systems.
Perplexity has defended the principle that customers should be able to use AI assistants to carry out tasks on their behalf. The opposing positions raise a difficult question: when an authorized customer directs software to use a website, whose access is legally relevant—the customer's or the AI provider's?
In March 2026, a federal district court granted Amazon a preliminary injunction restricting the disputed activity.
On August 4, the U.S. Court of Appeals for the Ninth Circuit vacated that preliminary injunction and returned the case for further proceedings.
The appellate court concluded that Amazon was unlikely to establish the necessary computer-access element of its federal claim on the record before it. Its reasoning distinguished actions performed by the customer through an AI tool from access performed independently by the AI company.
That decision was important for the immediate dispute, but it was not a final judgment establishing that AI agents have an unrestricted right to use third-party websites.
The litigation also involves questions of authorization, website policies, and technical design. The Verge reported on October 17 that Amazon had filed an amended complaint, indicating that the underlying conflict had not ended with the August ruling.
For AI developers, the case demonstrates that technical innovation alone cannot settle disagreements about platform access. Legal interpretation and website rules may determine which automated services can operate reliably.
Can Authorized Commerce Integrations Solve the Problem?
While some companies are restricting independent agents, others are developing controlled ways for AI systems to interact with their services.
Shopify is a prominent example. In January 2026, the company announced the Universal Commerce Protocol, developed with Google, to support structured interactions between AI shopping systems and merchants.
Its June 17 developer announcement expanded self-service access to agentic commerce tools, including the Catalog API and infrastructure supporting product discovery and checkout.
Rather than requiring an agent to interpret and navigate every page like a human shopper, structured integrations can provide product information and supported transaction capabilities through defined interfaces.
Shopify expanded that approach again in October. According to The Verge's October 14 report, the company announced broader support for shoppers connecting AI agents to its Shop platform, with access to product listings and purchases through Shop Pay. The reported arrangement requires buyers to approve purchases.
This represents a different approach from allowing unrestricted browser automation. Merchants retain defined transaction processes, while compatible agents receive a supported route to perform certain tasks.
Shopify's Universal Commerce Protocol also provides a common framework intended to reduce the need for separate integrations between every AI developer and every merchant.
Such systems are not a complete solution. Support still depends on participating services, compatible software, available capabilities, account authorization, and payment requirements. A shopping integration also cannot automatically provide access to unrelated activities such as airline reservations or bank-account management.
What Website Restrictions Mean for AI Users
For consumers, the immediate consequence is uncertainty over which tasks an AI assistant can actually finish.
An agent might successfully research products but fail when attempting checkout. Another might retrieve travel information but be unable to complete a booking. A third could encounter an account-security challenge that requires the user to take over.
These outcomes matter because the appeal of autonomous agents depends on completing useful workflows, not simply demonstrating that software can navigate websites.
Users also need to distinguish between an assistant that recommends an action, one that prepares a transaction for approval, and one authorized to execute the transaction. Those are different levels of automation with different security and accountability implications.
As AI systems develop more advanced capabilities, including the kinds of agent functionality discussed in NewsFusion365's coverage of Google's AI developments, dependable access to external services becomes increasingly important.
However, more sophisticated reasoning or better browser controls do not eliminate a website operator's ability to restrict automated access.
Security checks can also create friction for human users. If websites respond to increasingly capable bots with more aggressive verification systems, legitimate visitors may face additional challenges when logging in, shopping, or completing transactions.
The result could be a more fragmented experience in which AI assistants work smoothly with some services but require manual intervention—or cannot complete the requested task—on others.
What Comes Next for AI Agents and the Open Web?
The emerging divide is not simply between companies that support AI and those that oppose it. The more consequential question is how websites can recognize authorized agents, determine what those agents may do, and maintain appropriate controls over security and transactions.
Standardized integrations offer one potential path. Clearly defined permissions, user-approved transactions, and compatible authentication mechanisms could make automation more predictable without requiring websites to accept every automated visitor.
But wider adoption is not guaranteed. Businesses must decide whether supporting outside agents offers sufficient benefits, and developers must build systems that respect the technical and commercial conditions attached to access.
The next developments worth watching include the Amazon–Perplexity litigation, further changes to website access policies, and the expansion of supported agent-commerce integrations.
Ultimately, the future of AI agents will depend on more than how intelligently they can perform tasks. It will also depend on whether the services they need are willing—and technically prepared—to let them in.
0 Comments