Websites block AI agents: JPMorgan, Schwab shut them out
The web's uninvited guest is getting harder to miss — and this week the doors closed on the banking side too. What started as a handful of retailers turning away shopping agents has spread into a general posture: if you hold customer accounts or sell access to content, the default answer to an AI agent is no.
JPMorgan Chase and Charles Schwab have confirmed they block AI agents from accessing customer accounts, and The Verge's tally this week puts that pair at the head of a list that now spans airlines, retailers, job boards and travel sites. The banks' stated reason is security — an agent with your login is one prompt injection away from moving money — but the pattern across the whole list is broader than fraud prevention. This is the moment agent-blocking stopped being an edge case confined to Amazon's ongoing fight with Meta and became the industry's resting position, and it matters because the agent economy everyone is building toward assumes the opposite: that software will be welcome anywhere a person is. If account-holding institutions set the norm, every consumer agent demo shipped this year runs into a wall before it reaches a balance.
The range of who is doing the blocking is the story. TechCrunch, working from first-party statements, confirmed that Yelp denies non-human traffic unless the agent's owner pays for its data-licensing program, that Delta has no third-party agent integration and evaluates them "with security and the customer experience in mind," that United's terms of service bar "any robot, spider, other automatic device" without written permission, and that eBay has suspended real user accounts for agentic use. Google has reportedly suspended accounts linked to Meta's Muse assistant for "suspicious activity" and previously booted the agent startup Instinct out of Gmail. Cars.com and Amtrak are turning agents away on user reports. Even Cloudflare, which sells the infrastructure half of this, moved its crawler defaults in a way that broke Muse traffic for some sites in September — the blocks are now partly automatic, not just policy.
Amazon remains the most aggressive holdout — it blocks Muse, xAI's Grok, OpenAI's Dots and Instinct from shopping on users' behalf, and its lawsuit over Perplexity's agent access survived an appeals-court setback in August, with an amended complaint now pending. We covered Amazon's pivot from banning agents to selling ads inside its own in Amazon blocked AI agents for a year. Now it sells ads inside one, and nothing this week changes the underlying logic: the company that owns the storefront decides who gets to shop it.
The emerging alternative to refusal is a toll, and Yelp's stance is the clearest template: pay for a data license and your agent gets in, exactly the arrangement publishers spent the last three years negotiating with model developers over training data. Applied to live web access, that model turns every meaningful site into a metered endpoint — which is workable for a company with a sales team and a contract-signing habit, and closed to the individual builder who just wants their assistant to book a flight. The blocking wave is therefore also a pricing decision in disguise, and it lands hardest on the part of the agent ecosystem with the least leverage.
One practical complaint runs through all of it: none of these sites publish an agent policy a developer can read before shipping. The Cars.com and Amtrak blocks are known from user reports, Walmart's were diagnosed after the fact, and eBay users discovered the rules by having their accounts suspended. The barriers are cheap to erect — Cloudflare has offered a one-click crawler block since 2025, and its September default change did much of the work automatically — while the people building against them find out by failing.
The counter-case is coming from the companies whose agents are being turned away, and it is not a security argument — it is a customer argument. "Turning away a personal agent means turning away the customer behind it," Meta says, per The Verge, and Chris Canal, CEO of the commerce standards group EquiStamp, put the supply-side version bluntly: "If there are companies that are actively working to make that not happen, there are just fewer useful things the agents can do." Walmart, whose Muse blocks were traced to a human-verification interrupt rather than policy, is among a group — with Meta, Stripe, Sierra, Genesys, Rocket, NICE and Decagon — backing a new open standard for agent commerce, which reads as the industry building the polite doorway while the barricades go up everywhere else. Notably, the shopping partners that said yes to Muse — Gap, Sephora, Best Buy, Expedia, Wayfair, Ticketmaster, Shopify — are mostly brands that want the traffic, which is the oldest dynamic in retail: the ones who can afford to refuse, refuse.
The sharpest tension sits inside the banks themselves. Schwab is blocking third-party agents from customer accounts in the same season it wired Claude into a wealth-management mode alongside BlackRock — as we reported in Claude gets a wealth-management mode, wired into BlackRock and Schwab. That combination is the actual policy emerging from this wave: agents are welcome when they arrive through a signed integration with liability attached, and unwelcome when they arrive as a browser extension wearing someone's credentials. The security rationale is real — an autonomous agent sitting in an authenticated banking session is a genuine attack surface — but notice who bears the cost. It is not the enterprises that will negotiate API deals; it is the individual user whose weekend automation project now violates the terms of service, as eBay's suspended accounts already show.
What to watch: whether other account-holding industries — health systems, brokerages, payroll providers — adopt the banks' blanket refusal or their integration exception, and whether the agent-commerce standard becomes the toll gate that replaces today's ad hoc blocking. Amazon's amended complaint against Perplexity is the other thread to follow, since a ruling on agent access to a retailer's site would matter more than any terms-of-service update.
Have you hit a wall using an AI agent on a site you pay for — or does your bank's refusal sound like common sense? Tell us in the comments.
