The artificial intelligence revolution is transforming almost every aspect of our lives. Much of it rests on enormous AI models trained on vast quantities of information, from recipes, children's songs and philosophy books to newspaper articles, websites and social media posts. But the latest technological breakthrough promises something even more ambitious: AI agents that can navigate the internet and carry out tasks on our behalf.
These autonomous agents can search for information, order dinner, organize email inboxes and even purchase a new dress, often with little human involvement. Yet as technology companies race to turn this vision into reality, an unexpected obstacle is emerging. What happens when the very websites these agents depend on refuse to let them in?
Amazon, eBay, major financial institutions and other leading online services are increasingly restricting access to AI agents, setting up a confrontation that could reshape the future of online commerce. Behind the resistance lie fears of security breaches, fraud, unauthorized access to customer information and the loss of control over transactions.
Pulling the plug
According to a report in Wired, users of AI services developed by Meta, xAI, OpenAI and other major technology companies have reported encountering restrictions in recent weeks that prevent their agents from completing tasks on leading websites. The Verge has also cited U.S. analysts warning that such barriers could even affect the valuations of major AI companies, whose business models increasingly depend on the ability of autonomous agents to operate across the internet.
Yelp, the popular business review platform, has adopted a policy requiring AI agents to obtain access through its paid data licensing program. Delta Air Lines, meanwhile, restricts outside agents from booking flights on behalf of customers without authorization.
Financial institutions are taking an equally cautious approach. JPMorgan Chase and Charles Schwab have confirmed that they are blocking unauthorized AI agents from accessing customer accounts, citing security concerns.
Agents have also reportedly encountered restrictions on the websites of Amtrak, the U.S. passenger rail operator; Cars.com, a major American vehicle marketplace; and department store chain Kohl's.
eBay has explicitly prohibited automated agents from accessing its marketplace without prior permission. Its restrictions cover shopping bots and other automated systems attempting to place orders without human oversight.
Amazon has emerged as one of the most determined opponents of unauthorized AI shopping agents. The company began blocking Perplexity's shopping assistants last year and subsequently sued the AI developer, alleging that it had repeatedly asked the company to stop operating its agents on Amazon's website.
A federal district court initially sided with Amazon, granting a preliminary injunction restricting Perplexity's access. However, a federal appeals court overturned that injunction in August, ruling that Amazon was unlikely to prevail on the computer-access claims underlying its request. The decision did not end the lawsuit, and Amazon subsequently filed an amended complaint.
Perplexity maintains that its AI shopping assistant is effectively an extension of the person using it, arguing that customers should be able to delegate online tasks to software acting on their instructions.
Amazon, however, is now engaged in a similar confrontation with Meta over its new AI agent, Muse, which it says violates the retailer's website policies. Online reports also suggest that agents developed by other companies are encountering difficulties performing tasks on Amazon.
Amazon is refusing to give unauthorized AI shopping agents unrestricted access to its website
(Photo: REUTERS/Stephane Mahe)
For AI developers, the implications are significant. If agents cannot access major shopping platforms, banking services or travel websites, their ability to complete the everyday tasks they were designed to handle could be severely limited.
Meta has responded by emphasizing cooperation rather than confrontation. In a post published earlier this month addressing reports that websites were blocking Muse, the company argued that businesses should recognize the commercial interests they share with AI developers instead of immediately resorting to legal threats.
"Rejecting a personal agent means rejecting the customer behind it," Meta said. "We prefer working with businesses to address their underlying concerns rather than seeing them lose that customer."
Rather than pursuing a direct confrontation with Amazon, Meta has expanded cooperation with retailers including Walmart, Gap, Sephora and Best Buy, which have embraced approved ways for Muse to assist shoppers.
Shopify has also opened its network of online stores to AI-powered shopping tools, while television shopping networks QVC and HSN increasingly view purchases made through AI agents as a legitimate sales channel. Some retailers are even proactively supplying product catalogs to companies such as OpenAI and Google to ensure their merchandise can be discovered through AI assistants.
Other retailers are drawing more careful distinctions between allowing AI agents to browse their products and letting them complete transactions.
Tapestry, the parent company of Coach and Kate Spade, has restricted unauthorized agent purchases through its brand websites, in part because of concerns about fraudulent transactions. At the same time, it has embraced approved AI commerce channels: customers can purchase selected Coach and Kate Spade products directly through Google's Gemini and AI-powered search services, with explicit approval required for each transaction.
The differing approaches highlight a growing divide between businesses that view AI agents as an opportunity to reach customers through new channels and those concerned that autonomous systems could undermine security, bypass established commercial arrangements or weaken their direct relationships with shoppers.
Even websites that welcome AI agents can block them
Importantly, AI agents can encounter obstacles even when a website has not deliberately prohibited them.
Many online services operate sophisticated anti-bot systems designed to prevent hacking attempts, fraud, automated data collection and other suspicious activity. These defenses do not always distinguish between malicious automated software and legitimate AI assistants carrying out instructions from real customers.
According to user reports, Google has restricted certain accounts that used Meta's Muse after detecting what its systems classified as suspicious activity. Similar problems have reportedly occurred at Walmart, despite the retailer's willingness to cooperate with Meta's AI services.
The result is that an agent may be unable to complete a perfectly legitimate task simply because its automated behavior resembles that of a bot attempting to bypass a website's defenses.
Cloudflare, one of the world's major cybersecurity and internet infrastructure companies, is among the businesses developing tools to manage the growing volume of automated traffic, including AI-driven bots that place additional demands on websites and online services.
The challenge for website operators is to distinguish useful AI agents from unwanted automation without compromising security or opening the door to fraud.
For technology companies, the dilemma is equally difficult. They are developing increasingly capable systems that can understand and execute complex instructions, but those abilities become less useful if the services they need to access remain unavailable.
There is another question that could prove just as important: Are consumers actually ready to let artificial intelligence shop for them?
Although investors increasingly believe autonomous shopping will become a major growth market, current research suggests that many consumers remain reluctant to surrender control over purchasing decisions.
A survey by payments technology company NMI found that just 3% of American adults currently describe AI as a personal shopper capable of managing every stage of their shopping journey. Only 10% said they would be comfortable giving AI full control over a purchase, while 45% said they would not trust AI to make purchases on their behalf at all.
The findings suggest a distinction between using AI to research products or compare prices and authorizing it to spend money independently. Consumers may welcome help finding a bargain, but that does not necessarily mean they want software choosing what to buy, entering payment information and completing the transaction without their approval.
Roughly a third of active AI users surveyed in research cited by the report said they would never give autonomous agents the power to purchase or even order products for them.
Sky Canaves, a principal analyst at research company EMARKETER, told The Wall Street Journal that the technology industry may be misunderstanding consumers' attitudes toward shopping.
"People in technology circles mistakenly think consumers don't like shopping, but we actually enjoy going to the supermarket," she said. "Consumers still like making their own decisions and being in control, especially when money is involved."
That reluctance could complicate the ambitions of companies investing heavily in autonomous agents. The promise of software that can carry out almost any online task may be technologically appealing, but its commercial success depends on much more than the sophistication of the underlying AI models.
Businesses must decide which agents to trust, how to protect their customers and whether allowing outside software to complete transactions serves their interests. Consumers, meanwhile, must decide how much authority they are willing to hand over to machines.
For now, the future of autonomous AI agents rests on a fundamental contradiction: Technology companies are building assistants designed to navigate the entire internet, while some of its most important destinations are closing their doors.
And even if those doors eventually reopen, another question remains: Will consumers actually want to give their digital assistants the keys?