AI Agents Promised to Do Everything Online. The Web Is Locking Them Out

The major AI firms are marketing agents that book flights, make purchases and move money on your behalf. Yet an ever longer roster of websites is turning those agents away. The Verge has it that in the last few weeks JPMorgan Chase, Charles Schwab, Delta Air Lines, Yelp and Amazon have all put up roadblocks for the likes of Meta, OpenAI, Google, xAI and Instinct. It is the first true measure for Silicon Valley as to whether an agent's utility is limited by the doors that are opened to it.

The Blocklist Is Growing Faster Than the Agents

Those who make use of Meta's Muse, xAI's Grok or OpenAI's Dots and Instinct have found themselves turned away at a number of the web's better known sites. Yelp will only let in agents that pay for its data licensing program, Delta Air Lines is not allowing them to put in ticket orders and JPMorgan Chase and Charles Schwab have both put the word out that they block agents from customer accounts on security grounds. Some users say Cars.com and Amtrak no longer work with their agents as they once did, a sign that other firms are quietly doing the same.

The banks are the most forthright in explaining their position. "We don't want someone just giving their password to Muse, and letting Muse come in and take action on that client's behalf," said Rick Wurster, CEO of Charles Schwab, in an interview with Bloomberg. Bank of America's chief executive, Brian Moynihan, has stated his institution is putting the AI industry on notice to have a human authorize payments until the agents can be relied on to err less.

It is a far cry from what the makers of the agents had led one to believe. OpenAI, Meta and Google have been touting an assistant capable of any online task, be it emptying an inbox or purchasing shoes, with the understanding the rest of the internet would oblige.

Amazon Wrote the Playbook With Perplexity

This month's row over agents is not a new development. The trouble began last year when Amazon put a ban on Perplexity's agentic shopping and took the startup to court, contending that any company making purchases for a customer ought to do so in an open way and abide by a site's choice on participation. Perplexity's response was that Amazon has no desire for agents to be going about its store, one designed to put ads in front of human eyes, and put forward the view that its agent is no more than an extension of the user, akin to an assistant.

A judge made Perplexity stand down but an appeals court set aside the order in August; Amazon has put in an amended complaint since then. It is still to be determined in law if an agent working for a logged in user is to be considered the user or a bot.

The wall is coming up for the big players as well. Amazon has informed Meta that Muse is in breach of site policy with its ongoing access. Users say they are unable to get Grok, Dots and Instinct's agent to work on Amazon. Elon Musk retorted that perhaps he should go into competition with Amazon, which is telling of how little sway even xAI has with a store uninterested in its bot.

A Blocked Agent Is a Less Valuable Agent

Each closed door is a dent in the fundamental proposition of the agent business. "If there are companies that are actively working to make that not happen, there are just fewer useful things the agents can do," said Chris Canal, CEO of the AI safety evaluator EquiStamp, in an interview with The Verge. Canal makes the case that this has a direct bearing on the market's valuation of firms such as OpenAI and Anthropic, since their narrative is predicated on agents being present in all aspects of online life.

The resistance can be put down to economics. An agent will forgo the impulse buys and ad revenue that put money in a retailer's pocket. A bank is left with the liability if a bot puts funds in the wrong place. A review site sells its data and does not want it taken for free. There are good reasons for them to put their foot down and, at present, the law allows it.

Meta's Answer: Partner Where You Cannot Push Through

In lieu of a confrontation with Amazon, Meta has opted for the negotiating table. "Turning away a personal agent means turning away the customer behind it," the company wrote in a post earlier this month, adding that it would rather work with businesses to address their concerns than watch them lose that customer. The company has put pen to paper with Walmart, Gap, Sephora and Best Buy among others so that Muse can do business on their sites; Shopify too has made its storefronts available to agents.

As for the impediments their agents encounter, OpenAI and Google had no comment for The Verge, while Instinct would not be drawn on it. There is a price to be paid for such an approach: an agent which was meant to have free rein is now confined to those venues where an agreement has been reached.

The Bot Defenses Were Built for Scammers, Not Assistants

Sites that have not set out to ban agents are running into them all the same. The web is awash with anti bot systems and on Reddit there are reports of Google putting the brakes on some accounts for what it deems suspicious activity with Muse, or of checkouts at Walmart not going through in spite of the company's tie up with Meta. "These AI agents trigger those same safeguards," Canal said, noting that the defenses were designed to stop scams and hacking rather than assistants.

Cloudflare has put it within easy reach to take that kind of stance. In 2025 it rolled out a one click block on AI crawlers and from then on was blocking those that scrape for the purposes of search indexing as well as AI training, an effort to shield infrastructure from a deluge of bots. According to Canal, the web will be more of a nuisance for it, with more sites asking people to prove they are human, even a grocery site.

Will the Web Split Into Agent Friendly and Human Only?

In the years ahead the internet may be divided into two camps. Retailers and services that put in place agent partnerships, as the Walmart and Shopify merchants have with Meta, may become the places where agents do their shopping and booking by default. The likes of Amazon and the big banks might remain human only for now until matters of liability and revenue are put to rest. Should this come to pass, those with control over the agents could gain a real say in which stores get the traffic, a form of gatekeeping that regulators may examine at some point.

Founders could find their opening in between. A startup that can vouch for an agent's identity or show it is acting on behalf of a particular logged in user, or one that insures its transactions, could become the sort of infrastructure banks and retailers will require before they let agents in. Then there are the firms building agents from browser automation and scraping; they may see their wares dwindle with every defense update from a large site.

The first thing everyday users may notice is the friction. If anti bot systems keep catching assistants, there could be more identity checks and failed checkouts and accounts put on ice for suspicious activity even if the user has not employed an agent. As the Perplexity matter indicates, a court may at some stage make the call on whether the agent is the user, and such a ruling, if it comes, might carry more weight than any model release.

There are three things to watch: a clear precedent from Amazon's amended complaint, OpenAI and Google moving to formalize retail ties in the way Meta has, and rules from Cloudflare or the banks to allow a verified agent in. Any one of these could determine whether we get to the age of the agent via open doors or through a series of negotiated gates.

The Question Silicon Valley Unplugged Keeps Coming Back To

Silicon Valley Unplugged is the place for such questions as who is to be held liable when a machine does your work. It is the subject of debate at Sparknify's monthly live studio panel, where founders, investors and academics debate where technology is heading and how it changes daily life. The agent blockade serves to illustrate why these are important conversations. A product may be technically sound but can still come to naught if the businesses involved deem the risk too high; the operators and investors on the panel are the ones to spot that first.

In the episodes ahead, expect the argument to turn from the capabilities of agents to what the economy at large will countenance. The banks have made their position known for the time being, retailers are divided and the courts have yet to finish with the Perplexity matter. Sparknify will continue to have those voices in the studio so readers in the Bay Area and elsewhere can get both sides of the story before any rules are put down on paper.

In the end the agent era will not be won by the smartest model but by how many doors remain open to it.

Subscribe to Sparknify for Silicon Valley tech news and the people behind it.