In short
Major websites are increasingly blocking AI agents from shopping, booking, and browsing on users’ behalf. The standoff is forcing the AI industry to confront security, revenue, and access limits on the open web.
- Major sites are blocking or restricting AI agents from completing tasks online.
- Amazon’s fight with Perplexity has become a key legal test case.
- Meta is pursuing partnerships while other companies face outright blocks.
- Anti-bot systems and Cloudflare protections are also tripping up agents.
- The growing resistance could make the web more restrictive for both bots and humans.
Popular websites are increasingly blocking AI agents from doing tasks on users’ behalf, creating a growing clash between the companies building autonomous assistants and the services that host the web. In the past few weeks, major platforms and financial institutions have begun tightening access, raising a practical question for the AI industry: if agents cannot reliably enter the web, how useful can they really be?
The dispute has moved beyond theory. Users of agent tools from OpenAI, Meta, xAI, and Instinct have reported failures on major sites, while companies such as Yelp, Delta Air Lines, JPMorgan Chase, and Charles Schwab have acknowledged or effectively implemented restrictions. The result is an emerging tug-of-war over who controls digital access when software acts like a human user.
Why websites are blocking AI agents now
Web services are pushing back because AI agents can trigger the same security, fraud, and traffic controls built to stop bots, scrapers, and account abuse. In some cases, companies also see commercial reasons to limit automated access, especially when agents reduce the number of human visitors seeing ads or interacting directly with a platform’s interface.
The central problem is that an agent is not just a chatbot answering questions. It is software that can log in, navigate pages, click buttons, fill forms, compare options, and complete transactions. That makes it useful for customers — but potentially risky for the businesses whose systems it touches.
What is changing on the web?
What is changing is that more sites are no longer treating AI agents as ordinary user tools. Instead, they are increasingly classifying them as automated systems that may need to be blocked, licensed, or explicitly approved before they are allowed to interact with a service.
That shift has been visible across travel, retail, finance, and entertainment-adjacent platforms. Some companies are saying no outright, while others are allowing only licensed or partnered agents to proceed.
Which companies are drawing the line?
Several high-profile companies are now at the center of the debate because they either confirmed restrictions or were reported by users to be interfering with agent activity. The list includes both consumer-facing platforms and heavily regulated financial institutions.
| Company or service | Reported position on AI agents | Why it matters |
|---|---|---|
| Amazon | Has fought Perplexity’s agentic shopping and says third-party tools must respect its policies | A major test case for whether agents can shop across ecommerce sites |
| Yelp | Allows only agents covered by its data licensing program | Shows how licensing may become a gatekeeping model for access |
| Delta Air Lines | Prevents agents from booking tickets on behalf of users | Highlights travel as a sensitive category for automation |
| JPMorgan Chase | Blocks agent access over security concerns | Signals how cautious banks are about automated account access |
| Charles Schwab | Also blocks agent access due to security risks | Shows finance firms are treating agents as a potential threat surface |
| Cars.com and Amtrak | Users have reported new access problems | Suggests quieter, less public forms of blocking may be spreading |
How Amazon became the defining case
Amazon became the defining case because it was one of the first major web companies to openly challenge an AI agent that could complete purchases on behalf of a user. Its dispute with Perplexity has become a proxy battle over whether an AI assistant is simply a digital representative of the customer or an unauthorized third party operating on a platform.
Amazon argued that it had repeatedly asked Perplexity to stop letting its agent shop on the site. The company also maintained that automated services acting for customers should be transparent and should accept the service provider’s right to opt out. In Amazon’s view, that is not just a technical issue; it is a question of consent and control over platform participation.
Amazon’s position is that third-party systems making purchases for customers should operate openly and respect whether a service wants to take part.
Perplexity pushed back by saying the real reason for resistance was economic. The startup argued that Amazon benefits when human shoppers spend more time viewing ads and product placements, while an agent reduces the platform’s ability to monetize attention. The company also framed its agent as the user’s assistant rather than a separate actor.
Perplexity argued that an AI agent should be understood as an extension of the user, not as an outside intruder on the platform.
The legal fight has already produced multiple turns. A judge ordered Perplexity to stop its agents from shopping on Amazon’s site, but an appeals court later overturned that decision in August. Amazon has since filed an amended complaint, keeping the conflict alive and signaling that the legal framework for agentic commerce remains unsettled.
What do AI companies want these agents to do?
AI companies want agents to move from conversation into action. That means more than drafting messages or summarizing text. The ambition is for systems to manage practical online errands that people currently handle themselves, such as making reservations, managing inboxes, shopping, comparing prices, and booking travel.
That vision is a major part of the hype around autonomous agents from OpenAI, Meta, Google, xAI, and smaller startups. These companies have presented agents as a next generation of AI tools: software that does not simply assist from the sidelines, but actually performs tasks across the web.
Why the blockages matter for the industry
Why the blockages matter is that the practical value of agents depends on access. If they cannot reliably move through common websites, their usefulness shrinks from “do anything online for me” to a much narrower set of tasks.
That concern is now shaping investor and user expectations. If major services can quietly shut agents out, the promise of frictionless automation becomes harder to defend as a mass-market product.
What Meta, OpenAI, Google, and xAI are facing
OpenAI, Meta, Google, and xAI are all trying to build agent products in a web environment that is becoming less welcoming. They face the same basic problem as Perplexity: a service can advertise automation, but it still depends on thousands of independent websites choosing not to interfere.
Meta has been especially active in presenting a cooperative alternative. The company said after recent blocking reports that refusing a personal agent is effectively the same as turning away the customer behind it. Rather than fight every platform head-on, Meta has been building partnerships with retailers so its agent can operate where the retailer has agreed to allow it.
Meta has argued that blocking a personal agent is equivalent to blocking the customer using it, and that businesses should work with platforms to address concerns instead.
The company has partnered with retailers including Walmart, Gap, Sephora, and Best Buy, while Shopify has also opened access to its storefront network for AI agents. That approach suggests one possible future for agent commerce: a permission-based web where only approved partners can transact at scale.
OpenAI and Google have been less public in this specific dispute. The Verge reported that requests for comment did not receive immediate responses. xAI owner Elon Musk responded to the reports with a joke about possibly starting a rival to Amazon, underscoring just how competitive — and personal — the fight over agent access has become.
How anti-bot systems are complicating everything
How anti-bot systems are complicating everything is by making it harder to distinguish between helpful agents and abusive automation. Many websites already use defenses that are not designed specifically to stop AI assistants, but to prevent fraud, spam, credential stuffing, scraping, and other forms of abuse.
That means an AI agent trying to act like a person can get caught in the same net as malicious bots. In practice, this can produce confusing failures: an agent may suddenly be unable to log in, complete checkout, or access an account even on a website that has not publicly announced any special ban.
Users on Reddit have reported Google accounts being flagged for “suspicious activity” when used with Muse, Meta’s agentic tool. Other reports say transactions have failed on Walmart despite Meta’s partnership with the retailer. Those accounts may not prove a formal policy change, but they do show the friction that can emerge when agent traffic hits standard web defenses.
Security tools on the web are often built to stop programs that look like scammers or hackers, and AI agents can trip those alarms even when they are acting on a user’s behalf.
What counts as a bot, and who decides?
What counts as a bot is now a live business question, not just a technical one. Website operators are increasingly deciding that if software does the clicking, the typing, and the transaction work, then it may be treated as automated traffic no matter how human its intent appears.
That creates a legal and commercial gray area. If an AI assistant is simply the user’s digital proxy, should it be allowed in? If it increases fraud risk or reduces platform revenue, should it be denied? The web has not agreed on an answer.
Why Cloudflare matters in the wider fight
Cloudflare matters because it helps shape the infrastructure layer on which many websites rely. When Cloudflare changes how AI crawlers are handled, it affects not just one site but potentially a large portion of the web ecosystem.
In 2025, the company introduced a one-click option for blocking AI crawlers. It later expanded efforts against crawlers that both index search results and collect data for AI training. Cloudflare framed those changes as part of a broader attempt to reduce strain from AI traffic and to help websites maintain control over how their content is used.
The same tools that protect sites from scraping can also make life harder for legitimate AI agents. That tension is central to the current moment: the infrastructure meant to preserve the web’s openness is being used to decide which kinds of software are welcome and which are not.
Will AI agents make the web less usable?
Will AI agents make the web less usable? In some ways, yes — because the defenses created to stop them may also make everyday browsing more cumbersome for humans.
Chris Canal, chief executive of the AI safety evaluator EquiStamp, said the spread of blocking systems could reduce the value of AI agents by limiting the number of places they can actually work. If companies claim their tools can do nearly anything online but large platforms keep shutting them out, the promise becomes harder to sell.
Canal also warned that the response may end up making the web more annoying for ordinary users. More sites may introduce extra checks, step-up verification, and human confirmation screens that force users to prove they are not automated systems.
One likely outcome, Canal suggested, is a more cumbersome web with more checkpoints designed to separate people from bots.
That prospect matters beyond AI hype. If grocery sites, booking platforms, or shopping services harden themselves against agent traffic, users may face more interruptions just to complete routine actions online. The friction could spread far beyond the companies involved in the current disputes.
Why this fight may reshape online commerce
Why this fight may reshape online commerce is that AI agents challenge the core structure of how web businesses make money. Many online services rely on direct visits, ad impressions, behavioral data, or tightly controlled checkout systems. Agents disrupt all of that by turning browsing into delegated execution.
For retailers, that can mean less direct engagement and potentially less ad revenue. For travel and finance firms, it can mean a new and unfamiliar risk layer around identity, authorization, and transaction integrity. For AI developers, it means the success of their products depends on legal permissions and platform partnerships rather than just technical capability.
The likely outcome is not a simple winner or loser. Instead, the web may splinter into three layers:
- sites that explicitly welcome agents,
- sites that allow them only through licensing or partnerships, and
- sites that block them altogether.
That kind of segmentation would be a major departure from the universal, loosely open web that helped make earlier internet tools so powerful.
What happens next?
What happens next will likely depend on whether major AI firms can persuade more websites to opt in rather than resist. Partnerships with retailers suggest one path forward, but the legal fights and anti-bot defenses suggest the web will not become agent-friendly by default.
For now, the trend is clear: AI companies want their agents to act everywhere, while many services want tighter control over how, when, and whether those agents are allowed to operate. The result is an internet that may become increasingly designed not just for people, but for deciding whether the software standing in for them is welcome at all.
| Issue | AI company view | Website/operator view |
|---|---|---|
| Access to shopping sites | Agents should act as user assistants | Platforms can refuse automated purchasing tools |
| Security | Agents are legitimate user tools | Agents may resemble fraud or scraping bots |
| Revenue | Automation improves convenience and conversion | Automation may reduce ads, data collection, and engagement |
| Future of the web | Agents should work across most sites | Access may require licensing, partnerships, or outright blocks |
The fight over AI agents is no longer hypothetical. It is already playing out in browser sessions, login screens, checkout flows, and courtroom filings. The next phase will determine whether the agentic web becomes a mainstream convenience — or a patchwork of permissions, denials, and digital checkpoints.
Frequently asked questions
Why are websites blocking AI agents?
Websites are blocking AI agents because they can look like bots, trigger fraud defenses, and reduce direct human traffic. Some companies also want to protect ad revenue, control checkout systems, or limit access to data and services that are being used automatically.
Which companies are blocking AI agents?
Several major companies are involved, including Amazon, Yelp, Delta Air Lines, JPMorgan Chase, and Charles Schwab. Users have also reported problems on sites such as Cars.com and Amtrak, suggesting that other companies may be quietly tightening access as well.
What is the Amazon and Perplexity dispute about?
The Amazon and Perplexity dispute is about whether an AI agent can shop on behalf of a user inside Amazon’s store. Amazon says the agent violates its policies, while Perplexity argues it is simply acting as the user’s digital assistant.
Are AI agents still useful if websites block them?
AI agents are still useful, but their value drops if major websites refuse access. The more sites that block them, the less likely they are to reliably handle real-world tasks like shopping, booking travel, or managing accounts across the web.
Will websites become harder for humans to use because of AI agents?
Yes, that is possible. As websites add more anti-bot checks to stop AI agents, ordinary users may face more verification steps and CAPTCHA-style hurdles, making routine browsing and checkout processes more frustrating.
